Patent Grant
11983007
This present invention relates generally to equipment for applying roadway lane demarcation markings onto a roadway surface and, in particular, to equipment for maintaining pre-existing roadway lane marks by automatically applying new roadway marking material from a moving vehicle directly over and on top of a pre-existing roadway mark using machine learning network and machine vision technologies.
Existing roadway surfaces will typically include roadway lane demarcation markings to assist motorists for visually identifying lanes for controlling and directing traffic. In many cases the roadway markings are placed directly onto the roadway top surface and usually consist of paint of various colors such as white or yellow for asphalt-covered roadways, or white or yellow onto a black painted background to accentuate contrast between the roadway surface and mark (used especially on lighter-colored concrete roadways). In other cases, roadway markings may be placed in grooves previously milled into the roadway surface.
The visible contrast between the roadway surface and the roadway mark is an important consideration for drivers of vehicles to be able to quickly and reliably discern the correct traffic lane under both daytime and nighttime, dry and wet, roadway driving conditions.
Different materials have been perfected for roadway markings. The most common roadway marking material is in liquid form (generically called “paint”) which is sprayed onto the roadway surface from a forward-moving paint vehicle along a desired roadway mark path. The liquid material then dries, cures or solidifies forming a dry and semi-permanent roadway marking. The applied thickness of the liquid marking material may be approximately twenty thousandths of an inch (0.5 mm) but may also vary depending upon the roadway surface roughness, application specifications, and the type of marking material. Some common liquid marking materials include epoxy, polyurea, traffic paints, or other commonly used marking materials specifically formulated for the roadway marking industry. Examples of roadway marking material manufacturers include Sherwin-Williams of Baltimore, Maryland and Epoplex Inc. of Maple Shade, New Jersey.
Hot thermoplastic material is also commonly used as the roadway marking material. This material is first heated and melted, and then while in a liquefied, molten state, either sprayed (under pressure), ribbon extruded, or applied from a trough or shoe (screed) onto the roadway surface. Cold preformed sections (usually three feet or 91 cm in length) of thermoplastic can also be placed into position along the desired roadway mark path and then semi-liquefied with a torch. The semi-liquefied thermoplastic material first conforms to the roadway surface and then quickly cools and solidifies, binding to the roadway surface forming the desired roadway mark.
Another material used for roadway marks is supplied in tape form. The specially formulated roadway marking tape is the actual roadway mark in continuous flexible form having an adhesive coating on the bottom surface for affixing the roadway marking tape to the roadway surface. This product may be directly applied on the top surface of the roadway surface, inlaid directly into hot asphalt roadway surfaces, or preferably applied into a groove which has been previously milled into the roadway surface. An example of a continuous roadway marking tape product is Stamark™ Pavement Marking Tape manufactured by the 3M Company of St. Paul, Minnesota.
All the above roadway mark materials are effective for visibly defining the roadway lane demarcation marks during daytime and clear weather conditions but are less effective during wet and nighttime conditions.
To increase the wet and nighttime visibility of the roadway mark, reflective elements are applied onto the roadway mark material during the application process. These reflective elements reflect the on-coming headlights of an approaching vehicle back towards the approaching driver greatly improving nighttime visibility of the roadway mark. Reflective elements may be dispensed onto the top surface of either the freshly sprayed roadway marking or the melted thermoplastic to further improve the nighttime visibility of the roadway markings. Tape products may also integrate the reflective elements into the tape surface forming a composite reflective surface.
The most common reflective elements which are co-applied with the liquid or thermoplastic roadway marking material are small and generally spherically shaped glass beads. The beads are constructed so that they will efficiently back-reflect light from oncoming headlights thus self-illuminating the roadway mark. The term “retro-reflectivity” is used to describe this phenomenon.
For liquid-based sprayed materials, beads are commonly dispensed with a velocity equal and opposite that of the forward moving paint truck so that the beads fall almost vertically downward under the force of gravity and partially embed themselves onto the top surface of the freshly applied marking material. The top portion of the bead is exposed and thus able to retro-reflect the oncoming vehicle headlights.
The size of roadway beads varies but the most common bead diameters range from approximately fifteen to fifty thousandths of an inch (0.4 to 1.2 mm). A popular manufacturer of roadway marking glass beads is Potters Industries LLC of Malvern, Pennsylvania.
As the liquid marking material dries, cures, or in the case of liquefied (molten) thermoplastic, cools and solidifies, the beads are in effect glued and affixed to the marking material and hence to the roadway surface. The marking material (often referred to as the binder) along with the imbedded beads form a hard-composite structure having a rough raised and exposed top surface. The exposed portion of the bead above the cured binder captures and back reflects part of the on-coming light from vehicle headlights. The thickness of the finished roadway marking includes the thickness of the hardened binder along with the exposed portion of the bead above the cured liquid material surface.
Beads of two or more different diameters may also be simultaneously dispensed together to achieve different reflectivity properties for different environmental roadway conditions. Other non-spherical reflective elements may also be dispensed singularly or in combination with other types of reflective elements.
For example, twenty thousandths inch (0.5 mm) diameter beads may be co-dispensed with fifty thousandths inch (1.25 mm) diameter beads to better improve wet (rainy) nighttime roadway reflectivity. Dispensing two different types of reflective components is commonly referred to as a double drop process. Dispensing three types of reflective components (for example, two different sized beads and one irregularly sized reflective element) is referred to as a triple drop process.
A portion of the larger diameter bead may be able to protrude above the wet film thickness of the water on a wet roadway surface and still retro-reflect light from oncoming headlights. Because the larger diameter beads protrude significantly above the road surface, however, they are more susceptible to degradation over time. In contrast, the smaller diameter beads may be completely submerged under the wet film thickness of the water on a wet roadway surface and will not effectively capture, and therefore not back-reflect, the light from oncoming headlights, but because of their smaller size, are less susceptible to degradation over time. Other irregular shaped and dimensioned reflectivity elements are sometimes additionally dispensed along with beads to further improve roadway marking reflectivity under wet roadway conditions.
For roadway mark material supplied in tape form, the reflective elements are usually directly integrated into the tape and form a continuous composite structure. Some tape products have reflective elements that are imbedded in a polygon-shaped, raised profile arranged in a waffle-like pattern to capture and back-reflect the light from oncoming vehicle headlights for both dry and wet nighttime conditions.
All the above solutions are effective in producing initially high-contrast differentiation between the roadway marks and the roadway surface under different environmental driving conditions. This initially high contrast differentiation tends to degrade over time, however, for various reasons. For example, the actual binder material which defines the visible reflective shape of the roadway mark (usually rectangular shaped) may become worn with vehicular traffic and the passage of time, and may further become discolored because of lengthy exposure times to ultra-violet radiation from the sun. Further, black tire scuff marks may occur on the surface of the roadway mark further degrading mark visibility.
Differences in the coefficients of thermal expansion between the binder material and roadway surface may also cause a physical separation between them further degrading the ability to maintain a consistently visible roadway mark. Seasonal roadway temperature variations over time may cause cracking and peeling of the roadway mark further degrading the effectiveness of the marking.
The reflective elements for the liquid, thermoplastic, and tape applications which are installed along with their respective binder may wear away quickly as the result of frictional contact between the tires of passing traffic and the applied roadway mark. Roadway mark degradation also occurs as the result of the partial or complete scraping removal of the binder and its reflective beads or reflective elements from plows attempting to clear the roadway surface of snow in northern climates. Usually, the exact positions of the roadway lanes, and therefore the roadway marks defining those lanes, are not visible during the snow-plowing process thereby increasing the probability that the marks will be accidentally scraped and removed by the plow.
To maintain effective contrast between the roadway mark and roadway surface, roadway transportation agencies and others may periodically employ different roadway mark reapplication processes to maintain roadway mark contrast and visibility. One process first removes the worn pre-existing mark by specialized grinding machines (commonly called “grinding trucks”) or with a pressurized and circulating stream of water (commonly called “water blasting”). A new roadway mark is then reapplied onto the bare roadway surface at approximately the same position where the original mark existed. This particular reapplication process is costly and inefficient because it requires use of a dedicated and expensive piece of equipment (commonly referred to as a “grinding truck” or a “water blaster”) or a combination of marking-removal machines, to first remove the pre-existing worn lines, and then additional time and labor costs are required to again lay out a new roadway mark path and apply the new roadway marking. Furthermore, the motoring public is inconvenienced as the result of traffic lane closure for both the removal and the subsequent layout and reapplication of the roadway marks.
Another more popular and less expensive reapplication process applies new roadway mark material (for example paint and, if required, reflective elements) directly on top of the pre-existing roadway mark without removing the worn pre-existing roadway mark. This process of reapplying roadway mark material directly on top of the pre-existing roadway mark is commonly referred to as “maintenance striping,” and is the subject of this invention. Common maintenance striping processes may include a completely manually controlled process, or the completely manually controlled process may be partially automated. Both types of maintenance striping processes are now discussed.
For the completely manually controlled maintenance striping process, a driver first positions the paint truck along the pre-existing roadway mark path and then moves the truck forward in a longitudinal direction along the mark path at a speed consistent with the roadway mark material dispensing application process (approximately 10 mph). Having a clear frontal perspective view of the pre-existing roadway mark, the driver maintains the paint truck position and path substantially parallel with the pre-existing roadway mark path.
As the paint truck proceeds along the pre-existing roadway mark path, a first operator situated towards the rear of the paint truck and having an unobstructed view of the pre-existing roadway mark controls both the lateral position of the roadway mark material dispensing hardware (usually a moveable carriage more fully described below) and the dispensing process on one side of the paint truck (for example, the left side for painting a center line in the United States). A second operator may also be stationed towards the rear of the paint truck for controlling the roadway mark dispensing hardware and process on the opposite side of the paint truck (for example, the right side for painting a lane edge line). These operators are responsible for both properly aligning the material dispensing hardware over the pre-existing roadway mark and for controlling the timing of when to begin, and when to end, the dispensing of the roadway mark material.
For long lengths of solid line roadway mark dispensing (such as roadway center lines), the operators are primarily concerned with only aligning the dispensing hardware over the pre-existing mark and are not so concerned with continuously controlling the timing of when to begin, and when to end, the dispensing of the roadway mark material except at the beginning and end of the solid line mark. This is not the case, however, for repetitive skip line patterns where the operators must continuously align the dispensing hardware over the pre-existing roadway mark and simultaneously monitor and adjust the paint truck dispensing system to ensure that the newly applied marking accurately reproduces the existing skip line pattern by periodically manually correcting the timing of when to begin, and when to end, the dispensing of the roadway mark material.
Manual maintenance striping of a roadway mark therefore requires a driver to steer and align the paint truck along the pre-existing roadway mark path and at least one and possibly two operators for controlling both the lateral position of the roadway mark dispensing hardware and timing of when to begin, and when to end, the dispensing of the roadway mark material.
The dispensing hardware for liquid (or other types) of mark material (for example, an epoxy binder) usually consists of two laterally extendable and hydraulically controlled carriages mounted on opposite sides of the paint truck. Left and right-side carriages are usually provided so that the center and side lines may be painted (either singularly or simultaneously) as the paint truck moves along the roadway mark path and are controlled by their respective operators. The carriages further have attached and downwardly directed guns which spray the roadway mark binder material (for example, paint) onto the roadway surface. Reflective media dispensing nozzles may also be affixed to the carriage rearward of the roadway mark binder spray nozzles. Other roadway mark materials may require different dispensing hardware which is compatible with the roadway mark material.
The hydraulic control system for each carriage may include a conventional hydraulic steering control unit which cooperates with one or more hydraulically operable cylinders having a moveable piston (which is further attached to a piston rod) for laterally extending and retracting the carriage (one end of the cylinder is fixed to the paint truck body and the piston rod is attached to the carriage). The hydraulic steering control unit controls the direction of hydraulic fluid to the hydraulic responsive piston which in turn either laterally extends or retracts its respective carriage.
The operators manually control the lateral position and movement of the left and right-side mounted carriages (and therefore the liquid binder and bead nozzle lateral positions) by controllably rotating the hydraulic steering control unit via a conventional steering wheel. Turning the steering wheel in one direction may extend the carriage while turning the steering wheel in the opposite direction may retract the carriage, with the lateral velocity of the carriage determined by how quickly the operator can physically turn the steering wheel.
The hydraulic carriage control system allows the respective operators to laterally align the roadway mark material dispensing hardware over the pre-existing roadway mark path and to adjust the carriage position to account for slight positional inaccuracies of the driver in positioning the paint truck when attempting to follow the roadway mark path, especially around curved roadway mark paths.
Having aligned the nozzles at the correct lateral position with the pre-existing roadway mark path, the operator must then decide when to begin and when to end the dispensing of the roadway mark material as the paint truck proceeds along the pre-existing roadway mark path. Usually, the operator attempts to start dispensing mark material at the beginning edge of the pre-existing line segment. To accomplish this task, the operator must first visually locate the leading edge of the line segment and then estimate when to begin dispensing the roadway mark material taking into consideration the inherent turn-on delay of the dispensing valves and vehicle speed. Variations in both truck speed, dispensing valve turn-on delay, and operator response time usually result in positional misalignment between the actual beginning, or leading edge, of the pre-existing roadway mark line segment and the starting edge of the newly applied mark line segment. A positional alignment tolerance between the actual leading edge of the pre-existing roadway mark line segment and the starting edge of the newly applied mark line segment is usually allowed by most transportation agencies or others responsible for maintaining roadway markings. A mark edge positional tolerance may be plus or minus several inches (or centimeters) and can vary according to a particular transportation agency or other defined specification.
Having started dispensing the roadway mark material on top of the pre-existing mark, the operator continues to dispense the roadway mark material until the trailing edge of the pre-existing mark line segment comes into view, at which time the operator attempts to judge the proper time at which to stop the dispensing process. Trailing edge positional errors between the pre-existing mark and the newly applied mark may occur because of variations in truck speed, dispensing valve turn-off delays, and operator response time. Slower vehicle speeds may be necessary to give the operators more time to determine when to begin, and when to end, the dispensing of mark material thereby minimizing the leading and trailing mark edge positional alignment errors. A slower moving truck increases the amount of time necessary, however, to re-apply the mark material over the pre-existing mark.
Although successful in dispensing new roadway mark material on top of a pre-existing roadway mark, the manual process is prone to human error based upon the fact that the operator must simultaneously laterally align the dispensing hardware and control the timing when to begin and end the dispensing of mark material.
In addition, the operator cannot accurately determine the actual length of the line segment and will usually dispense new roadway mark material over the entire length of the visible pre-existing line segment without regard to the original line segment specification. For example, a common 15/40 skip-line pattern should include a 15-foot (450 cm) line segment with a 25-foot (750 cm) gap. Because of previous restriping activities, however, the 15-foot (450 cm) line segment may have been over sprayed and lengthened to a 17-foot (520 cm) line segment. With a manually controlling restriping process, the operator may overly dispense and restripe the entire 17-foot (520 cm) line segment even though the original specification calls for a 15-foot (450 cm) line segment, thus unnecessarily wasting a good percentage of roadway mark material. Alternately, the operator may short-dispense the line segment of a skip-line pattern. For example, if the required 15-foot (450 cm) paint line segment was previously under sprayed to a length of 13 feet (400 cm), the operator may only restripe the previous most recent and visible 13-foot (400 cm) line segment (the remaining 2-foot (60 cm) line segment of the original 15-foot (450 cm) line segment being worn away).
In an attempt to minimize the edge positional errors and to stripe the specified line segment length, “timer-based” controller systems have been created to assist the operator during the restriping process. These partially automated systems typically determine the distance travelled by the paint truck along the roadway mark path and use this information to appropriately turn-on and turn-off the paint nozzles (or other roadway mark material dispensing apparatus, including, for example, bead dispensing systems) to create the desired skip-line pattern. An example of a commercially available system for controlling the dispensing of roadway mark materials is model SM-5 previously manufactured by Skip-Line Inc. of La Grande, Oregon.
The distance travelled by the paint truck may be determined from a drive shaft-mounted aluminum ring having several permanent magnets imbedded around the outer circumference of the ring. As the drive shaft (and hence the attached ring) rotates, the spatially changing permanent magnetic flux is detected by a chassis-mounted conventional Hall-effect sensor or other magnetically responsive sensor which outputs a series of electrical pulses. For example, having twenty permanent magnets imbedded in the ring will produce twenty pulses for each drive shaft rotation. Because drive shaft rotation also causes rear wheel rotation via the rear axle and differential, the number of pulses produced at the drive shaft location will be proportional to the rear wheel rotation, and hence proportional to the linear distance travelled by the vehicle. The equivalent linear distance travelled per pulse is usually first determined by a calibration procedure before a roadway is restriped.
To calibrate the system, the driver will first physically measure a known length of roadway (for example, 1,000 feet or 300 m). The truck is then driven along this known length of roadway and the number of pulses produced by the magnetically responsive sensor is recorded by the timer. Knowing the distance travelled (in feet or other convenient length unit) and the number of corresponding pulses produced for this distance by the magnetically responsive sensor allows the system to calculate the equivalent linear distance (feet) travelled per pulse, and which may also be used to calculate vehicle speed in miles per hour (feet per pulse*pulses per second*3,600 seconds per hour*1 mile per 5,280 feet). The distance travelled by the paint truck may then be determined by counting the number of pulses and can therefore be used to determine an accurate line segment and gap lengths for dispensing the roadway mark material. Other technologies may be employed to determine the distance travelled by the paint truck and include a vehicle mounted conventional RTK enabled GPS like system.
For example, after completing the calibration procedure, one pulse from the magnetically responsive sensor may correspond to a travelled distance of 1.2 inches (30 mm). Assuming a 15/40 skip line is to be restriped, the controller would turn-on the valves to dispense roadway mark material for 150 pulses ((15 feet*12 inches per foot)/(1.2 inches per pulse)) and subsequently turn-off the valves thereby not dispensing the roadway mark material for 250 pulses. This turn-on and turn-off cycling of the dispensing valves would be repeated for the entire length of the skip line. The controller may also compensate for the turn-on and turn-off delay times of the dispensing valves, and other timing advance or delay issues.
Assuming the calibration distance to pulse ratio is constant throughout the restriping process, these types of controllers can repeat the painted line segment and gap lengths with good positional accuracy.
The distance to pulse ratio may change, however, during the restriping process. For example, it is well known that tire diameter is a function of tire pressure, and that tire pressure is a function of tire temperature. Variations in tire temperature can therefore cause changes in the diameter of the tire which subsequently changes the previously calibrated distance to pulse ratio. For example, increases in tire temperature during the restriping process may cause a change in tire pressure. This change in tire pressure may result in a change in tire diameter which may result in a distance error per tire revolution. Besides the change in temperature ultimately affecting the distance travelled per tire revolution, the operators may decide not to recalibrate the distance-to-pulse ratio before beginning a new painting application but instead rely upon previous distance-to-pulse values. Other factors may also affect the distance travelled per tire revolution such as tire wear and tire deflation caused by leaking or inoperable tire valves.
Errors caused by changes in the distance-to-pulse ratio during the restriping process, or by using previous and not current ratios, are cumulative and cause positional errors in the painted line segment and gap lengths. For example, assuming a 15/40 skip line is to be restriped and assuming an initial 1.2 inches (30.5 mm) per pulse ratio, a change from 1.2 inches (30.5 mm) per pulse to 1.25 inches (31.75 mm) per pulse would produce a skip line mark of 15.625 feet (476 cm) and not the desired 15-foot (457 cm) long skip line mark, a difference of 7.5 inches (19 cm). The gap length will also change from 25 feet (762 cm) to 26.04 feet (792 cm), a difference over one foot (30 cm). This dispensing cycle error is cumulative and continues throughout the restriping process, and if not quickly corrected results in an unacceptable restriped roadway mark pattern.
To adjust the dispensing cycle to account for slight variations in positional dispensing of the roadway mark material caused by the errors in the distance-to-pulse ratio or other errors, the operators visually observe the beginning position of where the roadway mark material is being dispensed and visually compare this position with the beginning position of the pre-existing roadway mark. If the start position of the dispensed roadway mark is not aligned with the start position of the pre-existing mark, the operators must manually lead (advance) or lag (delay) the timing of when to dispense the roadway mark material (commonly referred to as jogging) to realign subsequent start positions.
The accuracy of restriping exactly over the pre-existing roadway mark is greatly dependent upon the ability of the material-dispensing operators to both laterally align the carriage (and therefore the dispensing hardware, i.e., the paint-spraying nozzles) over the roadway mark as the paint truck moves along the roadway mark path, and further to advance or delay the dispensing starting position of the roadway mark material to account for any variations in the distance-to-pulse ratio and other factors. The positional accuracy of the partially automated restriping process is therefore dependent upon the accuracy and consistency of the calibration procedure and again on the judgment of the material-dispensing operators, and therefore is prone to errors.
Furthermore, the requirement that the operators have an unobstructed view of the roadway mark for both laterally aligning the carriage over the roadway mark and for advancing or delaying the timed dispensing cycle usually places the operators towards the rear of the paint vehicle and therefore in harm's way of high-speed passing or common lane traffic. Documented operator injuries have occurred because of collisions between the rear portion of the paint truck and passing or common lane traffic.
Previous attempts to completely automate the maintenance restriping of pre-existing roadway marks have particularly included systems which use the optical characteristics of the previously applied roadway mark material for controlling the lateral position of the mark material nozzles and the actual timing of when to begin and end the application of mark material.
For example, U.S. Pat. No. 3,101,175 issued to Carl F. Brown teaches a paint truck having a closed-circuit television system which is used to assist the paint truck driver in guiding the vehicle along the roadway mark path. The driver of the truck must carefully position the vehicle adjacent to the pre-existing roadway mark and, using the television image of the pre-existing roadway mark, attempt to steer the paint truck to continuously align the roadway mark dispensing hardware over the pre-existing roadway mark. The driver must continuously monitor the television receiver to maintain the dispensing hardware alignment over the pre-existing mark position in addition to controlling when to start and stop the dispensing of roadway mark material. Although this patent attempts to eliminate the operator from the rear of the paint truck, maintaining an accurate dispensing position over the pre-existing roadway mark is difficult because of the simultaneous tasks required of the driver to properly dispense the roadway mark material at exact times while maintaining both the truck and roadway material dispensing hardware alignment with the current roadway mark path, and also attempting to control the position of the paint truck along the pre-existing roadway mark path.
U.S. Pat. No. 3,229,660 issued to J. L. McLucas et al. teaches an apparatus for selectively applying roadway marking material to highway pavements and also for automatically controlling a paint-applying vehicle along a predetermined roadway mark path. Information-bearing signal elements placed beneath or on top of the roadway surface define a predetermined roadway mark pattern. The information-bearing elements may include strips of metal or radioactive material embedded into the roadway surface, or the previously affixed and optically responsive roadway mark material (paint). Detectors responsive to the respective information-bearing elements control dispensing new roadway mark material. A photocell is disclosed for detecting the presence of a painted roadway mark. A signal is generated when a roadway mark is detected which is then used to control the dispensing of mark material. The optical detectors work well assuming that there is sufficient roadway marking material available on the roadway surface for the photocell to optically distinguish between a marked and unmarked surface. This situation rarely occurs because traffic has diminished the optical distinguishing characteristics of the previously applied mark or has completely removed the mark from the roadway surface. The invention is therefore capable of applying mark material only at those positions where sufficient previously applied roadway mark material is currently optically detectable.
U.S. Pat. Nos. 5,054,959 and 5,169,262 issued to Wilson et al. teach a pavement line marking apparatus which comprises a support structure mounted to a moving paint truck having a marking detector (line scan imager) for detecting an old line marking, a transversely moveable and controllable paint applicator for depositing paint onto the old line marking, and a control system responsive to a signal from the marking detector to move the paint applicator into a position over the old line marking and to controllably deposit paint onto the old line marking. These apparatus work well if the old-line marking is detectable by the marking detector but fail if some of the old-line marking has been completely obliterated along the current roadway mark path. Further, if the beginning of the line mark is not clearly identified by the marking detector, the paint will not be deposited until the beginning of the line mark edge is clearly identified by the mark detector.
U.S. Pat. No. 5,203,923 issued to William H. Hartman teaches a control system for repainting old paint markings comprising a source of light which illuminates and electromagnetically stimulates the pre-existing roadway mark. A spectroscopic detector analyzes the spectral content of the reflected light from the pre-existing roadway mark to determine the presence or absence of known preselected chemical constituents of the mark material for both controlling roadway mark material dispensing and for tracking the roadway mark path. Reliable detection of the roadway mark requires, however, that the electromagnetic spectral emission response of the chemical constituents of the roadway mark material be matched with the wavelength of the illumination to achieve the greatest amount of stimulated (fluorescence) spectral emission, and if the mark is worn away by traffic it no longer provides stimulated emission.
U.S. Pat. No. 5,456,548 issued to Smyrk et al. discloses an apparatus for applying lines of pre-existing roadway mark configurations onto a roadway or pavement surface and to accurately repeat the roadway mark patterns. The apparatus comprises a survey system mounted near the front of the paint vehicle having a roadway mark detector (a charge-coupled device, or “CCD,” line scanning imager) to transversely scan the roadway surface, and a pattern transition detector taught by a neural network to recognize line pattern changes, and in response thereof, control the dispensing of mark material to accurately repeat the line pattern changes. The accuracy of the apparatus in determining the exact point at which a transition occurs between the current line pattern and a following line pattern (for example a mark and skip pattern), and therefore the dispensing of roadway mark material, depends upon how well the neural network is able to learn from the various pattern changes. The neural network disclosed does not (a) determine the location of the roadway mark with respect to the paint nozzles or (b) control the movement of the paint carriage to place the nozzles over the old-line markings.
An apparatus to fully automate the restriping process will (1) reliably detect the location of the roadway mark with respect to the paint vehicle irrespective of roadway illumination conditions, shadows, color changes of the roadway surface, intersections, imager field of view variation, blending of the roadway mark into the roadway surface, and background clutter and noise; (2) automatically and accurately align the roadway mark material dispensing nozzle(s) at the beginning edge location of the first mark of the first striping cycle and the beginning edge locations of subsequent marks throughout the restriping process; (3) automatically and accurately dispense roadway mark material over the pre-existing roadway mark(s); (4) accurately maintain mark and gap lengths for each skip-line cycle; (5) automatically dispense roadway mark material (including the binder material and reflective components) for single, double, or shadow (contrast) line applications; (6) monitor the dispensing process of roadway mark materials; (7) be easily installed and retro-fitted to existing line-striping vehicles, and particularly for line-striping vehicles having a manually controlled hydraulically operated carriage positioning system; (8) automatically determine the desired skip-line pattern; (9) automatically self-calibrate the distance-to-pulse ratio during the restriping process; (10) reduce errors in determining the distance-to-pulse ratio; and (11) improve the start and ending positional alignment between the newly applied and pre-existing roadway marks, with or without the pre-existing roadway mark being optically detectable. None of the prior art addresses all these requirements. Other improvements will become apparent in view of the present invention.
Although current maintenance striping technology using the optical characteristics of the previously applied roadway mark material for controlling the dispensing of roadway mark material has been partially successful using classical machine vision and other technologies, further improvements to the prior art may be made to more fully automate and increase the accuracy and speed of the restriping process.
These improvements include using machine learning network and machine vision-based technologies. The machine learning network-based technologies may include fully connected conventional neural networks, convolutional neural networks (CNNs) and the like, and/or support vector machines (SVMs) among others.
Machine learning network technology has been shown to mitigate the problem of reliably detecting the location of the roadway mark within an actual roadway surface image, i.e., a roadway image having shadows, color changes, missing portions of the roadway markings, poor or varying roadway illumination, background clutter and image noise. In particular, fast responding convolutional neural networks are well suited for accurately detecting and locating the position of the roadway markings within an image irrespective of changes in roadway environmental conditions.
Thus, there is a need in the roadway marking industry for a roadway mark maintenance striping apparatus that requires less manual labor, increases operational safety for the operators, is more accurate and efficient and less expensive than the current roadway maintenance striping technology available today.
To meet the needs identified above and others which will be apparent from a review of the current maintenance striping technology and in view of its purposes, the present invention provides a new and improved apparatus and method for accurately applying new line marking material over pre-existing roadway marks regardless of the condition of the current roadway marks and without the need for an operator to control the maintenance striping process using machine learning network and machine vision technologies to accurately detect and locate the roadway marking.
To further overcome the shortcomings of current roadway mark maintenance restriping technology, a new apparatus and method for placing marks over pre-existing marks along a roadway mark path are provided. A basic object of the invention is to provide an improved apparatus for automatically repainting or otherwise replicating existing traffic lane demarcation lines on roadway or other surfaces.
Another object of the invention is to provide an improved apparatus for automatically repainting or otherwise replicating partially obliterated traffic lane demarcation lines on roadway or other surfaces.
Another object of the invention is to provide an improved apparatus for automatically repainting or otherwise replicating completely obliterated non-visible traffic lane demarcation line segments on roadway or other surfaces.
A further object of the invention is to provide an improved apparatus to automatically and accurately align the roadway mark dispensing hardware over the pre-existing roadway mark(s).
Yet another objective of the invention is to provide an improved apparatus to automatically and accurately align the roadway mark dispensing hardware over pre-existing solid single or double roadway mark(s).
And yet another object of the invention is to provide an improved apparatus to automatically and accurately align the roadway mark dispensing hardware over pre-existing skip line single or double roadway mark(s).
Another object of the invention is to provide a machine vision and machine learning network based control system for aligning the roadway mark material dispensing hardware directly over the pre-existing solid roadway mark.
Another object of the invention is to provide a machine vision and machine learning network based control apparatus for aligning the roadway mark material dispensing hardware directly over the pre-existing solid or skip-line roadway mark.
And yet another object of the invention is to provide an improved apparatus for accurately maintaining mark and gap lengths for skip-line marks.
Still another object of the invention is to provide an improved apparatus to provide an accurate start and end positional alignment between the newly applied and pre-existing roadway marks.
A further object of the invention is to provide an apparatus which is easily installed on existing roadway or other mark striping vehicles.
A further object of the invention is to provide an apparatus which is easily installed on existing roadway or other mark striping vehicles having a conventional hydraulically controlled paint carriage.
Another object of the invention is to provide an apparatus which automatically and dynamically calibrates the distance traveled by a striping vehicle.
Another object of the invention is to provide an apparatus which automatically and dynamically calibrates the distance traveled by a striping vehicle using machine learning network and machine vision technologies.
Another object of the invention is to provide an apparatus which automatically and dynamically calibrates the distance traveled by a striping vehicle using GPS and the like technology.
Another object of the invention is to provide an apparatus which automatically and dynamically calibrates the distance traveled by a striping vehicle using real time kinematics (RTK) enabled GPS and the like technology.
Yet another object of the invention is to provide an apparatus which computes a roadway mark path.
Another object of the invention is to provide an apparatus which computes a roadway mark path projection.
Another object of the invention is to provide an apparatus which computes a roadway mark path projection using machine learning network and machine vision technologies.
Yet another object of the invention is to provide an apparatus which computes a roadway mark path projection using machine vision and convolutional based neural network technologies.
Another object of the invention is to provide an apparatus which detects a roadway mark path using machine vision and fast segmentation convolutional neural network (“Fast-SCNN”) technologies.
Another object of the invention is to provide an apparatus which computes a roadway mark path using machine vision and fast segmentation convolutional neural network (“Fast-SCNN”) technologies.
Another object of the invention is to provide apparatus for aligning the roadway mark material dispensing hardware directly over the pre-existing solid or skip-line roadway mark using machine vision and fast segmentation convolutional neural network (“Fast-SCNN”) technologies.
Other objects will become apparent in view of the present invention.
The present invention includes a control system for positioning a marker over a pre-existing roadway surface mark. In the first embodiment of the control system, a calibrated imager images the roadway surface including a pre-existing roadway mark. A sensor determines the lateral location of the mark. A computer is responsive to the imager and is programmatically configured to: (a) process the image using a machine learning network and (b) produce an error signal based upon the lateral location difference between the processed image of the pre-existing roadway surface mark and the marker. A replacement mark is then laid over the pre-existing roadway mark using an actuator attached to the marker configured to be responsive to the error signal and position the marker over the pre-existing roadway mark. To position the marker over the pre-existing roadway mark, the actuator may comprise a motor, such as an electric motor, that controls a hydraulic steering unit configured to laterally position the marker over the pre-existing roadway mark.
In certain embodiments, the error signal is produced based upon the lateral location difference between (a) the lateral distance of the pre-existing roadway surface mark with respect to the vehicle and determined from the roadway mark image, and (b) the current lateral location of the marker with respect to the vehicle determined by a linear position sensor. The image is then processed using a fast segmentation convolutional neural network. The actuator then positions the marker over the pre-existing roadway mark.
In other embodiments of the control system, an electromagnetic radiation source, such as a laser or laser line generator, is attached to the marker for producing a line pattern on the roadway surface. A calibrated imager images the roadway surface including the pre-existing roadway mark and the line pattern. A computer is responsive to the imager for producing an error signal based upon the lateral location difference between (a) the lateral distance of the pre-existing roadway surface mark determined from the image of the roadway surface mark, and (b) the current lateral location of the marker with respect to the vehicle determined from the image of the line pattern. The image is again processed using a fast segmentation convolutional neural network. The actuator remains responsive to the error signal for positioning the marker over the pre-existing roadway mark. In other embodiments the imager may be selectively responsive to the electromagnetic radiation source, for example, the imager may comprise a band pass optical filter responsive to the electromagnetic radiation source.
The imager may be synchronously triggered directly using a deterministic timing source or asynchronously triggered with the images time tagged with a deterministic timing source. In certain embodiments the time deterministic signals produced by the deterministic timing source are periodic signals.
The image of the roadway (which includes the roadway mark) is processed by a supervised machine learning network system, such as a convolutional neural network, to mitigate the problem of reliably detecting the location of the roadway mark within an image having various roadway illumination conditions, shadows, color changes of the roadway surface, intersections, imager field of view variation, blending of the roadway mark into the roadway surface, and background clutter and noise.
The present invention further provides the ability to predict the location of the roadway marking underneath the carriage and nozzle locations.
The present invention further includes an apparatus for restriping a pre-existing substantially rectangular roadway mark. The roadway mark has been previously placed on a roadway surface along a roadway mark path. The apparatus comprises a vehicle for moving along the roadway mark path having a roadway marker for dispensing roadway mark material onto the pre-existing roadway mark and a sensor for determining the lateral location of the marker. An imager mounted on the vehicle is downwardly focused onto the roadway surface for imaging the roadway marks. A computer system responsive to the imager and configured to (a) process the image using a machine learning network and (b) produce an error signal based upon the lateral location difference between the processed image of the pre-existing roadway surface mark and the marker. Finally, an actuator, attached to the marker and responsive to the error signal, positions the marker over the pre-existing roadway mark.
The present invention still further includes an apparatus for replicating pre-existing skip line roadway marks. The apparatus includes a moving vehicle. An imager with a field of view for imaging the roadway surface mark is affixed to the vehicle. A marker positioned on the vehicle for dispensing roadway mark material in wired or wireless communication with a sensor configured to determine the lateral location of the marker, is also affixed to the vehicle. The apparatus further comprises a computer responsive to the imager and the sensor and programmatically configured to (a) process the roadway mark surface image using a machine learning network, (b) produce an error signal based upon the lateral location difference between the processed image of the pre-existing roadway surface mark and the marker, and (c) produce a dispensing signal from the processed image for dispensing roadway mark material over the pre-existing roadway skip line marks. Finally, an actuator responsive to the error signal and configured to positioning the marker over the pre-existing roadway mark is attached to the marker.
It is to be understood that both the foregoing general description and the following detailed description are exemplary, but are not restrictive, of the invention.
The invention is best understood from the following detailed description when read in connection with the accompanying drawing. It is emphasized that, according to common practice, the various features of the drawing are not to scale. On the contrary, the dimensions of the various features are arbitrarily expanded or reduced for clarity. Included in the drawing are the following figures:
The present invention provides machine learning network based roadway marking systems used for repainting or otherwise replicating existing roadway traffic lane demarcation lines on roadway surfaces, a process commonly referred to as maintenance restriping.
Referring now to the drawing, in which like reference numbers refer to like elements throughout the various figures that comprise the drawing,
Center skip line 12 usually follows the longitudinal directed center of the roadway 1. A roadway mark path 16 defines the path which center skip line 12 follows, and the longitudinal center line of center skip line 12 is coincident with roadway mark path 16. Roadway mark path 16 is shown as a dashed line on roadway surface 4, and edge lines 10 and 14 are usually offset a given distance in the lateral direction from roadway mark path 16 and are therefore substantially parallel to center skip line 12. It is understood that roadway mark path 16 is not visible on the roadway surface 4 but only illustrates and indicates the longitudinal center line of center skip line 12. Other roadway marks may be offset from roadway mark path 16.
Usually, roadway lane edge lines 10 and 14 are continuous lines but may have breaks or otherwise segments which are not marked. Roadway traffic exit lanes are good examples of where the edge lines 10 and 14 may have breaks or may otherwise not be parallel with roadway mark path 16. Likewise, center skip line 12 could be a single solid line, or a double solid, a single or double skip line, or a combination of these or other lines.
Center skip line 12 comprises a cyclic pattern of roadway line mark segment 18 followed by an unmarked gap segment 20. This mark and gap segments cycle is repeated continuously on roadway surface 4 along roadway mark path 16 but may change depending upon the roadway mark specifications. For example, the center skip line pattern may change to a solid single or double line or even a roadway mark comprising one solid line mark parallel to a skip line, such as a conventional roadway passing mark. The invention is not limited to the particular type of center or edge line patterns and includes solid single- and double-line patterns, single and double skip-line patterns, other patterns or various combinations of line patterns.
Center skip line 12 has cyclic length 22 with mark segment 18 having length 24 and gap segment 20 having length 26. Skip line patterns may be noted as two numbers separated by a delimiter, the first number indicating the mark segment length 24 followed by the second number which indicates cyclic length 22. For example, a 15/40 (the delimiter is the ‘/’) skip line pattern defines mark segment 18 length 24 of 15 feet (450 cm) and cyclic length 22 of 40 feet (1,200 cm), yielding a computed gap segment 20 length 26 of 25 feet (750 cm). Many other skip line patterns exist and may include 10/40, etc. Also, skip line patterns may be expressed in metric units (meters).
A conventional paint vehicle 50 having a right-handed Cartesian coordinate system 52 is further shown moving in a forward longitudinal direction 28 within lane 6 and along roadway mark path 16, restriping the roadway mark line segments of center skip line 12. It is understood that the term “vehicle” is given its broadest meaning, including any conveyance, motorized device, or moving piece of mechanical equipment for transporting passengers or apparatus. More specific and preferred examples of vehicles are trucks and road marking machines.
As indicated in
Roadway mark segments are usually characterized by rectangular shaped marks defined by near and far longitudinal edge lines and beginning and ending lateral edge lines. For example, mark segment 18 is substantially rectangular having near longitudinal edge line 40 (i.e., the longitudinal edge line closest to vehicle 50) and far longitudinal edge line 44 (i.e., the longitudinal edge line farthest from vehicle 50), and beginning lateral edge line 42 (i.e., the first lateral edge line approached by vehicle 50 traveling in direction 28) and ending lateral edge line 46 (i.e., the second lateral edge line approached by vehicle 50). The edge lines form a substantially rectangular-shaped boundary of the roadway mark 18. Lateral edge lines 42 and 46 define the beginning and ending lines, respectively, of mark segment 18, and points 43 and 47 define the center points of lateral edge lines 42 and 46, respectively.
Further shown in
Referring additionally to
The lateral distance between paint guns 84 and 86 may be manually adjusted to accommodate the restriping of parallel solid or skip double lines (for example, a solid line parallel to a skip line, usually used to designate an allowed passing zone, or two solid lines, usually used to designate a no passing zone, etc.). In a similar fashion, the lateral distance between reflective bead guns 88 and 90 can be manually adjusted to allow lateral alignment with paint guns 84 and 86, respectively.
Referring additionally to
As carriage 80 is laterally extended and retracted, the vertical projection of the nozzle movement of the paint guns 84 and 86 onto surface 4 produces a lateral projection line 81 (shown in
A conventional laser line generator 102 is shown mounted underneath the frame (which includes front, side, and rear frame members 116, 118, and 120, respectively) of carriage 80. Laser line generator 102 produces a fanned pattern of laser light 104 downwardly focused onto, and intersecting with, roadway surface 4 thereby producing a line pattern 106. The fanned pattern of laser light 104 is further characterized by a fan angle 108. Line generator 102 may also include infrared and ultra-violet lasers, or other focusable electromagnetic radiation sources. Line pattern 106 is positioned such that it is within area 70 and is imaged by imaging system 60. Imaging system 60 may then determine the lateral position of carriage 80 (imaging system 60 has been previously calibrated), and thus further determines the lateral positions of paint guns 84 and 86 and their respective nozzles and also bead guns 88 and 90 (lateral offset(s) adjusted). Imaging system 60 includes a CCD or CMOS camera 252 having an optical bandpass filter 262 which passes the reflected light from the roadway surface 4 including the mark segment 18 and line pattern 106.
Thus, as carriage 80 moves in a lateral direction inward to and outward from vehicle 50, line pattern 106 also moves giving a visual indication (for a visible laser line generator) of the lateral position of carriage 80 and which is subsequently imaged by camera 252. The lateral positions of the paint guns 84 and 86 (and their respective nozzles) and bead guns 88 and 90 are therefore also visually indicated by line pattern 106 taking into consideration any fixed offsets between the paint and bead guns and laser line pattern 106. Laser line generator 102 may also be moved laterally along the frame of carriage 80 and positioned so that line pattern 106 is laterally aligned with one of the paint guns, for example paint gun 84.
Laser line generator 102 may also be mounted to frame 54 projecting the fanned pattern of laser light 104 first horizontally with respect to surface 4 and then subsequently reflected downward by a mirror mounted on carriage 80 again forming line pattern 106 with roadway surface 4. Carriage 80 may further have reflective ruler markings 115 placed onto front frame member 116 of carriage 80, which may be imaged by imaging system 60 and which then may also indicate the lateral position of carriage 80. Laser line generator 102 may also include a conventional laser pointer projecting a substantially circular “spot” pattern onto roadway surface 4 and within imaged area 70.
Also alternately attached to frame 54 is a conventional draw wire sensor 110 (shown hidden as a dashed outline in
A laterally moveable paint carriage 130 identical to carriage 80 is attached to the passenger side of vehicle 50 and is shown in a slightly extended position beyond the passenger's side of vehicle 50 in
Further attached to carriage 130 are laterally extendible cylindrically shaped support arms 144 and 146 (not shown). A hydraulic piston 148 (not shown) is positioned between support arms 144 and 146. The moveable end of hydraulic piston 148 is attached to the side frame member of carriage 130, and the other end of the piston is secured to frame 54 of paint vehicle 50. Hydraulically powering piston 148 provides the necessary force to laterally extend or retract carriage 130 from paint vehicle 50 thereby enabling the positioning of paint guns 134 and 136 along with their respective nozzles and respective bead guns 138 and 140 over a roadway mark.
Carriage 130 further has a laser line generator or laser pointer mounted to its frame for projecting a laser line onto roadway surface 4 within imaged area 75, reflective ruler markings on the front frame member, and a draw wire sensor or other transducers for determining the lateral position of carriage 130.
Imaged area 70 includes any pre-existing roadway 1 center skip line 12 (or any other center line which may include single or double solid, or a combination of a skip and a solid line, or any combination thereof) with vehicle 50 travelling anywhere within lane 6. Similarly, imaged area 75 includes any pre-existing roadway 1 edge line 10 with vehicle 50 travelling anywhere within lane 6. Both imaged areas 70 and 75 laterally extend past the full lateral extension of their respective carriages 80 and 130, and also image their respective roadway surface 4 laser line patterns or spot images and/or carriage ruler markings.
As shown in
Antenna 152 receives satellite GPS signals 156. A communication cable 158 electrically connects antenna 152 to receiver 154. GPS receiver 154 along with IMU 155 may cooperate together to form a conventional inertial navigation system (INS).
Imaging system 60 may also be mounted over carriage 80 on a fixable swingable mount (i.e., the mount can swing back along the side of vehicle 50 when not required) having a frontal field of view projected forward in the longitudinal direction and so positioned to image area 70 including the roadway mark segment 18 and the line pattern 106. Imaging system 65 may be similarly mounted over carriage 130 to image area 75 including the edge line 10 and its respective laser line pattern.
Other locations on vehicle 50 for mounting imaging systems 60 and 65 for imaging areas 70 and 75, respectively, are possible. For example, the imaging systems 60 and 65 may be mounted backwards of the cab of vehicle 50 and having a rearward field of view. The imaging systems 60 and 65 may also be mounted on the paint carriages 80 and 130 respectively, or a combination of mounting positions may be used.
Referring now to
Outlet ports 186, 188, 190, and 192 of valves 178, 180, 182, and 184 connect to the near ends of flexible conduits 194, 196, 198, and 200, respectively. The far ends of flexible conduits 194 and 196 are connected to paint guns 84 and 86, respectively, and the far ends of flexible conduits 198 and 200 are connected to bead dispensing guns 88 and 90, respectively.
In response to pressurized air flow, the respective paint and bead guns open permitting the pressurized paint and/or beads to be forcibly dispensed onto roadway surface 4. The flexible conduits allow delivery of air to the paint and bead guns as the carriage laterally moves to align the guns (and their respective nozzles) with the pre-existing roadway mark. The material supply lines to the individual paint guns 84 and 86 and the individual bead guns 88 and 90 are not shown to avoid clutter and add clarity in
Solenoid valves 178, 180, 182, and 184 each have separate positive and negative electrical connections for supplying electrical energy to activate their respective valve switching solenoids. Fused electrical power is supplied to the positive terminals of valves 178, 180, 182, and 184 via terminals 202, 204, 206, and 208, respectively. Fused electrical power to operate the valve solenoids may be derived from a 12-volt battery (not shown) of vehicle 50.
The negative terminal of valve 178 connects via a line 210 to one terminal of an electronically controlled switch 212. The other end of switch 212 connects to ground via a line 214.
The on-off state of switch 212 is controlled by an externally generated electrical control signal which flows to the control terminal C via a control line 216. In a similar fashion, the negative terminal of valves 180, 182, and 184 connect via lines 218, 220, and 222 to one terminal of electronically controlled switches 224, 226, and 228, respectively. The other ends of switches 224, 226, and 228 connect to ground via lines 230, 232, and 234, respectively. Similarly, the on-off state of switches 224, 226, and 228 are controlled by an externally generated electrical signal which flows to their respective control terminals C via control lines 236, 238, and 240, respectively.
A computer 703 (shown in
In addition, each valve 178, 180, 182, and 184 has conventional protective circuitry 242, 244, 246, and 248 (circuitry details are not shown), respectively. The protective circuitry minimizes any generated fly back voltages induced across the respective positive and negative solenoid voltage terminals during solenoid-initiated valve switching.
A similarly constructed roadway mark material pressurized air control system 250 (shown in
A manufacturer of commonly used solenoid valves for controlling the distribution of pressurized air to control the flow of paint and/or beads through their respective guns is MAC Valves, Inc. located in Wixom, Michigan.
Referring now to
Mount 254 enables camera 252 to be independently rotated about the three axes 268, 270, and 272 in directions 274, 276, and 278, respectively, which enables camera 252 to be spatially positioned to image area 70. Mount 254 is further affixed to a vertical leg 280 of angle bracket 258 via conventional bolts 282, 284, 286, and 288.
Plexiglas globe 256 encloses both camera 252 and mount 254 and further has a distal hemispherical surface 290 and proximal mounting lip 292. Lip 292 has a rectangular-shaped groove 294 for accepting an O-ring 296 and additionally provides surface area 298 for mounting globe 256, via circumferentially arrayed conventional bolts 300, 302, 304, 306, 308, 310, 312, and 314, onto vertical leg 280 of bracket 258. The horizontal leg 281 of right-angle bracket 258 is affixed to the top surface of a compressor enclosure 56 (see
Imaging system 65 is identical to imaging system 60, having a camera or imager 330, a 3-axis adjustable mount 332, a protective Plexiglas globe 334, and an angle bracket 336 (all not shown, including the parts of those components). Imager 330 is identical to imager or camera 252 having a lens element 338, an optical filter 340, and an optical axis 342. Power, data, and control signals communicate with imager 330 via an electrical cable 344 (all not shown).
Referring now to
Pump 402, reservoir 404, filter 406, relief valve 408, steering control unit 410, and cylinder 411 are in fluid communication with each other via conduits 418, 420, 422, 424, 426, 428, and 430. One directional hydraulic fluid flow is indicated by single-headed arrows (for example, single headed arrow 432) while doubly connected headed arrows indicate possible hydraulic flow in both directions (for example, doubly connected headed arrows 434 and 436). A manufacturer of steering control units is Eaton Corporation of Beachwood, Ohio.
A splined shaft 415 (not shown) of steering control unit 410 is axially aligned with and is attached to a shaft 412 of electric motor 414 via a connecting hub 417. Steering wheel 416 is axially aligned with and is also attached to shaft 412 of electric motor 414 via hub 417 with conventional circumferentially mounted bolts 419a, 419b, and 419c.
Internal to electric motor 414 is a programmable motor controller 413 which externally communicates via a communication bus or cable 421 with computer 703 (shown in
Electrical power is supplied to motor 414 via a power cable 423 and may be derived from the 12-volt battery of vehicle 50 or another electrical power source. Motor 414 may be a conventional direct drive permanent magnet synchronous motor (PMSM), having high torque and low operational rotational velocity. Motor 414 is further adapted to be easily installed and retrofitted onto the steering control unit 410 without requiring special modifications to either steering wheel 416, steering control unit 410, or a support stand 425 (support stand is shown in
Reservoir 404 connects to the inlet port 438 of pump 402 via conduit 418. Outlet port 439 of pump 402 connects to the pressure (P) port 446 of steering control unit 410 and the input port of relief valve 408 via conduit 424. The output port of relief valve 408 connects to reservoir 404 via conduit 422. The tank (T) port 441 of steering control unit 410 connects to the inlet port of filter 406 via conduit 430. The output port of filter 406 connects to reservoir 404 via conduit 420. The right port (R) 442 of steering unit 410 connects to the port 444 of cylinder 411 via conduit 426, and the left port (L) 443 of steering unit 410 connects to the port 445 of cylinder 411 via conduit 428.
Cylinder 411 has piston 440 having a connected piston rod 447 which extends and retracts in directions 452 and 453, respectively, in response to hydraulic fluid flow in conduits 426 and 428. The proximal end of piston rod 447 connects to piston 440 and the distal end of rod 447 attaches to the inside of side frame member 118 of carriage 80 at attachment point 96 using a clevis fastener 448. Thus, hydraulically extending rod 447 laterally extends carriage 80 and hydraulically retracting rod 447 laterally retracts carriage 80.
A clockwise rotation of splined shaft 415 of steering control unit 410, either produced automatically by motor 414 or manually with steering wheel 416, causes a pressure differential between the surface areas of piston 440. This pressure differential forces piston 440, and therefore piston rod 447, to move into hydraulic cylinder 411 in direction 453, thereby laterally retracting carriage 80 into the driver's side of vehicle 50.
A counterclockwise rotation of the splined shaft 415 of steering control unit 410, either produced automatically by motor 414 or manually with steering wheel 416, causes a pressure differential between the surface areas of piston 440. This pressure differential forces piston 440, and therefore piston rod 447, to extend outwardly from hydraulic cylinder 411 in direction 452, thereby laterally extending carriage 80 outwardly from the driver's side of vehicle 50.
It is therefore understood that computer 703 may communicate with motor 414 via commands sent to controller 413 via cable 421, and therefore may control the lateral position of carriage 80. Electrically disengaging motor 414 (defined as allowing the free rotation of shaft 412) by computer 703 allows spline shaft 415 of steering control unit 410 to be manually rotated via steering wheel 416 without any interference or assistance from motor 414. With a disengaged motor 414, the lateral position of carriage 80 may be manually controlled as if motor 414 had not been inserted into hydraulic steering system 400.
Hydraulic steering system 450 (not shown) controls the lateral movement of carriage 130. Hydraulic steering system 450 is identical in every respect to hydraulic system 400 except that the hydraulic connections are reversed on the hydraulic cylinder so that a counter clockwise rotation of the steering wheel (or motor) retracts, and a clockwise rotation of the steering wheel (or motor), extends carriage 130. Other hydraulic systems and other configurations are possible for controlling the movement of carriages 80 and 130.
Referring to
The cylinder 411 may be further equipped with a linear position sensor 110 (as shown in
Referring to
Drive shaft 508 is further connected to a conventional rear axle differential which in turn drives the rear axle of vehicle 50. Further attached to the rear axle are driver and passenger side rear wheels 57 (see
As drive shaft 508 rotates in the direction 510, collars 502 and 504 along with spacer 506 also rotate in the same direction 510. Cylindrically shaped permanent magnets 512 and 514 are imbedded and potted within, and are radially arrayed around the outer circumference of, collars 502 and 504, respectively. Further, collar 504 is rotatably displaced from collar 502 so that magnets 514 are radially aligned between magnets 512. A manufacturer of these types of magnetic shaft collars is Electro-Sensors, Inc. of Minnetonka, Minnesota 55343.
Conventional Hall-effect sensors 516 and 518 are positioned in close proximity to the outer circumference of shaft collars 502 and 504, respectively, and are attached to the body frame 54 of vehicle 50 by conventional mounts (not shown). Sensors 516 and 518 detect the changing magnetic flux produced by magnets 512 and 514, respectively, as collars 502 and 504 rotate in response to rotation in the direction 510 of drive shaft 508.
In response to the changing magnet flux, sensors 516 and 518 produce active low signals 520 and 522 (represented by pulses 524 and 526, respectively, illustrated along a time or “t” axis in
In response to input signals 520 and 522, NAND gate 532 produces an active high output signal 534. Output signal 534 from NAND gate 532 flows onto a line 536 and is an input signal to bus interface circuit 538. NAND gate 532 is shown separately for clarity but may be directly incorporated into interface circuit 538.
Signal 534 is composed of the signals from shaft collars 502 and 504. Having collar 504 rotatably displaced from collar 502 allows twice as many magnetic pulses than that possible from just one collar given a particular shaft collar size and number of magnets per collar. Having additional collar 504 increases the angular resolution of drive shaft 508 rotation per pulse. More collars rotatably displaced from one another may be added to increase the angular resolution of drive shaft 508, or multiple sensors may be radially displaced from one another and mounted circumferentially around a one or more collars.
For example, if collar 502 has a total of 36 magnets, then each active low pulse 524 corresponds to an angular rotation resolution of 10 degrees. With second collar 504 also having 36 magnets and rotatably displaced so that magnets 514 are aligned between magnets 512 of collar 502, a second non-interfering active low pulse 526 is produced between pulses 524, in effect giving an angular rotation resolution of 5 degrees. Therefore, each pulse of signal 534 corresponds to a known angular rotation of drive shaft 508 and therefore a known angular rotation of rear wheel 57.
The longitudinal distance travelled (or the longitudinal distance that will be travelled) by vehicle 50 is then easily determined by counting the number of pulses of signal 534 and multiplying this number by the distance travelled per pulse of signal 534. This distance travelled per pulse value in the past was prone to a multitude of errors as previously mentioned in the background section of this document and is greatly diminished according to the preferred embodiment of this invention. More than one Hall-effect sensor may be constructively arranged to produce additional pulses.
Interface circuit 538 may incorporate a conventional microprocessor 540 in bi-directional communication with bus interface circuitry 542. Interface circuitry 542 handles all bi-directional communication to and from local bus 544 to microprocessor 540. Microprocessor 540 may input signal 534 from line 536 and be programed by computer 703 to perform computational tasks such as counting a certain number of pulses of signal 534 over a particular time interval via a conventional gating signal. For example, computer 703 may communicate to interface circuit 538 a “start count” command which would instruct circuit 538 to begin counting the pulses of signal 534, and then communicate to interface circuit 538 a “stop count” command which would instruct circuit 538 to stop counting the pulses of signal 534. Computer 703 may then request the total pulse count of signal 534 which occurred between the “start count” and “stop count” commands from circuit 538, whereby circuit 538 would send the total pulse count back to computer 703 via local bus 544, or the pulse count of signal 534 may be synchronously or asynchronously sent to computer 703.
Bus interface circuitry 542 conditions microprocessor 540 signals intended to be sent onto bus 544 to be compatible with the chosen bus 544 specification, and conditions signals received from bus 544 intended to be sent to microprocessor 540 to be compatible with the signal specifications of microprocessor 540. Bus 544 may include, for example, conventional CANopen or EIA-485 (formally referred to as RS-485) communication protocol specifications. Thus, interface circuitry 542 is in bi-directional communication with computer 703 (and other components shown in
Referring now to
An identical machine vision and machine learning computer control system 750 controls carriage 130 but is not shown, except that system 750 will not include a corresponding GPS receiver, IMU and GPS antenna (only a single GPS receiver 154, a single IMU 155, and a single GPS antenna 152 is required). Additionally, only a single drive shaft positional sensor 500 is required for vehicle 50 and interfaces to both control systems 700 and 750.
Computer 703 is in bidirectional communication with a similar computer 752 (not shown) of identical machine learning based control system 750 via a bi-directional bus 710. Alternately, the tasks performed by computer 752 may be managed entirely by computer 703.
Computer 703 is in bi-directional communication (i.e., sends and receives data) among various components, including GPS receiver 154 and IMU 155, imager 252, pressurized air control system 460 valve 427, drive shaft positional sensor 500, hydraulic valve 427, linear position sensor 110 and precision time protocol (PTP) network timing unit 111 via master bus 712 and local busses 724, 266, 714, 544, 421, 716 and 113 respectively, and each component is in bi-directional communication with each other.
Master bus 712 may be composed of a number of different individual local busses, each individual local bus having different electrical and mechanical specifications supporting their respective communication specifications. Bus 712 may also provide power. For example, local bus 266 may be a camera link bus, USB compatible or ethernet bus, local bus 421 may be a CANopen bus and local bus 113 may be an ethernet bus and, when grouped or bundled together form part of master bus 712.
Keyboard 706 and liquid crystal display 704 (or similar display technologies) are conventional computer peripherals and are connected to computer 703 via bidirectional universal serial buses (USB) 718 and 720, respectively. Keyboard 706 allows an operator to enter alpha-numeric and other data into computer 703 and display 704 displays information from computer 703 for viewing by the operator. Display 704 may also be a conventional “touch” display allowing the operator to both view information and enter data by selectively touching areas displayed on the display 704, similar to the displays used on “smart” cell phones such as the Apple 11 phone. In addition, a conventional computer-compatible mouse and joystick are also provided (not shown) for entering data into computer 703 by the operator.
Power supply 708 supplies electrical power to all components shown in
GPS antenna 152 is electrically in communication with GPS receiver 154 via electrical cable 158. Antenna 152 and receiver 154 are adapted to receive conventional GPS signals 156 from any GPS satellite system (or a general GNSS, for example, the Russians' GLONOSS system), or from a GPS or GPS-like pseudolite system. In addition, receiver 154 is further adapted to use RTK data via a separate communication channel (not shown) to compliment the satellite-derived GPS (or GNSS) data thereby increasing the calculated GPS positional accuracy of vehicle 50.
The single (master) antenna/receiver GPS system shown in
An inertial measurement system (IMU) 155 may also be combined with the GPS receiver 154 to complete an inertial navigation system (INS). Typical INS systems include a first and a second GPS antennas and receivers and can be used to determine heading information along with roll, pitch and yaw information of paint truck 50. An example of an INS system using a first and second GPS antennas (and receivers) is model number n580 manufactured by Honeywell Aerospace.
GPS receiver 154 decodes signals 156 received by antenna 152 and uses RTK data (via the separate communication channel) to determine the geographical location (longitude, latitude, and altitude, or the ECEF position) of antenna 152. The location of antenna 152 is known with respect to coordinate system 52 through calibration.
Bi-directional communication with GPS receiver 154 among the other components of system 700 is via local bus 724 and master bus 712. As previously stated, only one RTK enabled GPS system is required on vehicle 50 for accurately measuring travelled distance of vehicle 50.
Imager 252 is a conventional progressive scan complimentary metal oxide semiconductor (CMOS) imager for capturing an image 253 of a roadway surface having a sensor with a rectangular-shaped pixel array usually arranged in a rectangular format. The pixel array converts light into electrical signals. An example of an imager is a GigE compatible, model number acA1440-73gc, color camera manufactured by Basler AG of Ahrensburg, Germany.
For example, the sensor for the acA1440-73gc has an array of 1440×1080 active pixels and conforms to the GigE specification. It is noted that more than one imager may be incorporated into control system 700.
Attached to imager 252 is conventional lens 260 which may have optical bandpass filter 262 (shown in
Included within imager 252 is electronic circuitry (not shown) which communicates status, control, and image data using a conventional ethernet bus interface via local bus 266 and master bus 712 to computer 703. Further, imager 252 may be triggered to acquire an image from a trigger signal derived directly from GPS receiver 154 through computer 703 (via an image acquisition system 726) or from other time-deterministic trigger sources (i.e., the time of occurrence of the trigger signal is known), or acquire images asynchronously, in which case the images are time-stamped using PTP timing unit 111, computer 703, or by other means.
Linear position sensor (transducer) 110 measures the relative lateral linear displacement of carriage 80 with respect to frame 54 of vehicle 50. For example, linear position sensor 110 may be a conventional industrial digital CANopen draw wire sensor model number WDS-5000-P115 manufactured by Micro-Epsilon of Raleigh, North Carolina (United States office) having the sensor housing mounted on frame 54 of vehicle 50 and a flexible steel (Teflon-coated) wire affixed to side frame member 118 at attachment point 96 of carriage 80.
Linear sensor 110 may also be a conventional laser range finder affixed to frame 54 and focused on a reflective target mounted on the inside of side frame member 118 of carriage 80, or may be a conventional linear variable differential transformer (LVDT).
The cylinder 411 may also incorporate an internal linear displacement sensor 110 (refer to
Position and other data are communicated between sensor 110 and computer 703 via local bus 716, which then becomes a member of master bus 712. Computer 703 may poll (request) sensor 110 for positional information or sensor 110 may continuously send positional data to computer 703. The position of carriage 80 is known via sensor 110 with respect to coordinate system 52 (offset adjusted) using conventional calibration techniques.
Thus, it is understood that the relative lateral positional movement of carriage 80 with respect to frame 54 is determined by sensor 110, and relative distances moved by carriage 80 can be calculated from differences in position locations, as well as position locations (and distances) of objects mounted on carriage 80, including the positions of paint and bead guns and their respective nozzles. The position of carriage 80 with respect to the coordinate system 52 (offset corrected) may also be determined using conventional calibration techniques.
Pressurized air control system 460 is in communication with computer 703 via local bus 714 and master bus 712. Thus computer 703 can control the dispensing of roadway mark material via pressurized air control system 460.
Computer 703 also has internally available peripheral component interconnect (PCI) expansion slots and/or peripheral component interconnect express (PCIe) expansion slots. For example, computer 703 may be provided with a conventional PCIe input-output (I/O) board inserted into a PCIe compatible expansion slot for sending and receiving digital control signals from computer 703 to external peripherals, such as conventional roadway mark material pressurized air control system 460, and for receiving digital signals from external peripherals to computer 703. Additionally, computer 703 includes a graphics processing unit (GPU) 713 such as that offered by NVIDIA model number TITAN V which is PCIe slot compatible. Other types of expansion slots may be provided to accommodate additional peripheral cards.
Computer 703 further includes an image acquisition system 726 for hardware interfacing imager 252 with computer 703. Acquisition system 726 may include a conventional frame grabber PCIe expansion slot compatible image frame grabber card, such as model number NI PCIe-1433, or a high-performance camera link frame grabber card manufactured by National Instruments Corporation of Austin, Texas. For GigE ethernet based imagers, a frame grabber card is not required.
System 726 may also include a random-access memory (RAM) buffer for storing acquired images 253 from imager 252, and handles all of the software overhead (control, image data transfers, etc.) for interfacing imager 252 to computer 703. Acquisition system 726 further has an external image trigger input 728.
In response to an external trigger signal placed onto input 728, acquisition system 726 sends a control signal to imager 252 via busses 712 and 266 to acquire or otherwise “snap” an image 253 at a known time. Image 253 data (pixel intensity (grayscale or color) and pixel array location information) are then transferred from imager 252 to the on-board buffer memory of acquisition system 726 via the respective busses and then subsequently transferred to data memory 806 (shown in
Acquisition system 726 may also respond to software instructions to acquire image data from imager 252 at known times. Thus, image data may be acquired in response to hardware or software-initiated trigger signals whose time of occurrence is known, or computer 703 may receive asynchronous images 253 data from imager 252 and time-stamp the acquired images. Additionally, some imagers are able to store image data directly into memory 806 bypassing the on-board buffer memory of acquisition system 726.
Computer 703 also includes conventional timing module 730 which may be programmed either by computer 703 or from an external programming source via signals placed on line 732 to perform certain timing tasks, and may be used as a trigger source for acquiring images from imager 252 at known times or to provide a time-stamp for asynchronously acquired images.
An external trigger source (not shown) may generate and accurately control the timing of the external trigger signal and may be programmed by computer 703 to produce various trigger signals. For example, the trigger source may be programmed to generate a periodic trigger signal having a known frequency. In response to the periodic trigger signal, imager 252 acquires a sequence of images 253 having accurate and known time intervals between each acquired image. A sequence of images 253 may then be acquired in response to a deterministic external trigger signal.
The trigger source may be a conventional programmable signal generator, or may be derived from the computer 703 internal timer, a timing module 730, an external microcontroller-based system or GPS receiver 154.
Images 253 may also be obtained asynchronously from imager 252 and subsequently time-stamped using an external precision time protocol (PTP) network timing unit 111. An example of a PTP timing unit 111 is model number TM2000A manufactured by TimeMachines, Inc. of Lincoln, Nebraska. Computer 703 is configured to communicate with timing unit 111 via bus 712 and local bus 113.
Therefore, it is understood that a sequence of images may be acquired by imager 252 and placed into data memory 806 (see
The acquired image is stored in memory 806 as an array of grayscale values for black and white images or as a three-dimensional array for color images having a one-to-one correspondence with the pixel array. For example, a black and white CMOS sensor having a 640×480 pixel array will output a 640×480 array of grayscale values for each image. Alternately, a color imager having a 640×480-pixel array would have, for example, three 640×480 arrays to accommodate red, green and blue (commonly referred to as RBG) image color values.
Referring to
Operating system software 802 may include a real time operating system (RTOS), UNIX, LINUX, Windows (offered by Microsoft Corporation), or other compatible operating system software, and performs conventional operating system software functions and is capable of executing various programs stored in program memory 804 of computer 703.
Program memory 804 includes an image correction program 808, a pixel-to-distance program 810, an image analysis program 812, a mark path projection program 814, a machine vision carriage control program 816, a dynamic positional calibration program 818, and a dispensing control program 820.
Image correction program 808 inputs raw image data acquired from camera 252 and subsequently corrects the raw image data for optical pin-cushion or barrel distortion produced by lens 260 (and possibly Plexiglas globe 256) and then secondly corrects for perspective distortion using a conventional homography algorithm. Both the raw image and corrected image data are stored in data memory 806.
For example, the object space (i.e., the actual physical field of view of camera 252) of area 70 includes substantially rectangular-shaped roadway mark segment 18, having near longitudinal edge line 40 (i.e., the longitudinal edge line closest to vehicle 50) and far longitudinal edge line 44 (i.e., the longitudinal edge line farthest from vehicle 50), and beginning lateral edge line 42 (i.e., the first lateral edge line approached by vehicle 50) and ending lateral edge line 46 (see
The edge lines form a substantially rectangular-shaped boundary roadway mark segment 18. Lateral edge lines 42 and 46 are commonly referred to as the beginning and ending lines of mark segment 18, respectively. The rectangular-shaped roadway mark boundary is clearly defined by substantially “straight” lines in object space.
Referring now to
Referring to
Also shown in both
The lateral position of the carriage 80 with respect to the vehicle coordinate system 52 may also be determined from the linear position sensor 110 (calibrated and offset corrected).
The respective distortion parameters required by image correction program 808 to correct for optical distortion are determined by a conventional optical distortion correction program, such as offered by MathWorks, Inc. of Natick, Massachusetts, and which is known in the camera calibration art. In addition, perspective distortion is then corrected using a homography transformation (it is assumed that the roadway surface 4 is planar to the imager sensor of camera 252).
Image data of dimensionally defined 2-D checker-board patterns are used by image correction program 808 to determine the correction data necessary to minimize the optical and perspective distortions, along with the appropriate software. Moreover, the image u-v coordinates may extend beyond the actual undistorted image boundaries, again assuming the roadway surface 4 is planar within the field of view of camera 252, and in particular includes the area under the paint and bead guns and their respective nozzles. The calibration and correction data is stored in memory 806
Pixel-to-distance transformation program 810 transforms each pixel of the undistorted image into an equivalent object space distance (for example, one pixel in image space in the u-axis direction of the corrected image may correspond to 0.25 inches or 6 mm in object space in the x-direction), or transforms each object space coordinate into a corresponding undistorted image space coordinate. Further, the x-y-z object space location of each pixel is determined and referenced to coordinate system 52 by conventional calibration methods.
In particular, pixel-to-distance transformation program 810 may determine the object space x-y-z coordinates of the image space origin 905 with respect to the vehicle coordinate system 52 of the undistorted image from which all other pixel coordinates in object space may be determined. Data necessary for performing this transformation are again experimentally determined from known object space x-axis and y-axis dimensions of an imaged checker-board pattern, and determining the z-axis coordinate of the roadway surface 4. Thus, every pixel (in image space) has an equivalent object space x-y-z axis coordinate referenced to coordinate system 52. These pixel-to-object and object-to-pixel distance transformation data are then stored in data memory 806.
The pixel-to-distance transformation program 810 may also transform array space data to object space data and vice versa. Array space data is discussed in reference to
Referring now to
Within image analysis program 812 is a supervised machine learning network 815 which inputs the corrected image 914. Machine learning networks may include fully or partially connected conventional neural networks, convolutional neural networks (CNNs), fast segmentation convolutional neural networks (Fast-SCNN), or support vector machines (SVMs). All the above-mentioned machine learning networks are well known in the computer vision art.
The machine learning network 815 is configured to classify each pixel of image 914 and outputs a classification array 817 for the image 914. The size of the classification array 817 is identical to the size of the pixel array of image 914, and each element of the classification array 817 has a one-to-one correspondence to each pixel of image 914.
For supervised convolutional neural networks, the classification array 817 gives a probability (i.e., confidence) having a range [0,1] that a certain pixel belongs to the classification. One or more classification schemes may be used. For example, one classification scheme may define the confidence level for a pixel representing a roadway mark, and another classification scheme may define the confidence level for a pixel representing an unmarked roadway surface.
Another classification scheme may define the confidence level for a pixel imaging a yellow roadway mark or a white roadway mark if a color imager is used, such as the aforementioned Basler camera. The classification schemes are “learned” during training of the machine learning network using a training data set (labelled data).
For training machine learning networks for roadway mark applications, a training data set comprising roadway mark image data is used. One source of roadway mark data is that provided by CARLA, a simulator for producing realistic roadway marks (see “CARLA: An Open Urban Driving Simulator” by Dosovitskiy et. al., 1st Conference on Robot Learning, Mountain View, California). Using the training set data, the neural network automatically adjusts internal parameters (commonly referred to as weights and biases) to minimize a cost function.
The image analysis program 812 further has a programmable filtering program 829 which operates on the classification array 817 confidence values 827 and sets the respective confidence values to a 1 or 0 depending upon a programmable threshold value 831.
The output 833 of the filtering program 829 is therefore an array of 1s and 0s mapped to the corresponding pixels of the image pixel array and the classification array, with the ‘1’s representing the specified classification and meeting the threshold criteria (programmable threshold value 831).
For example, if a pixel contained within an imaged roadway mark has a classification array 817 confidence value of 0.98, and the programmed threshold value is 0.90, the filtering program 829 would set the corresponding pixel in the output data array 833 equal to ‘1’.
As an example, and referring to
The image data of
The programmable filtering program 829 then compares the classification probability of each pixel to the programmed threshold 831. If the classification probability is less than the programmed threshold 831, the programmable filtering program 829 sets the value of the respective pixel to 0 in the output data array 833. If the classification probability is equal to or greater than the programmed threshold 831, the programmable filtering program 829 sets the respective pixel to 1 in the output data array 833.
For a well-trained machine learning network 815, the pixels 834A representing the roadway blotch 871 may have a low probability assigned to them as belonging to a roadway mark 867. The output data array 833 may then be represented as shown in
For a poorly trained machine learning network 815, the pixels 834A representing the roadway blotch 871 may have a higher probability assigned to them as belonging to a roadway mark 867. The output data array 833 may then be represented as shown in
Depending upon the level of training of the machine learning network, the other pixels groupings 834B 834C, 834D 834E and 834F may or may not be properly classified. Note that the location of each element of array 833 follows conventional array labeling (row, column)
The output 833 of the filtering program 829 may flow to the conventional image segmentation program 835. Depending upon the desired level of specificity, the segmentation program 835 may restore data 833 and may output array 836 sharply defining the boundary 837 of the imaged roadway mark representation as illustrated in
Segmentation program 835 may also fill in image area gaps, such as pixel groupings 834E and 834F. It is understood that the pixel array size of 24×24 is for illustrative purposes only and is not limiting to this disclosure.
Each element of array 836 maps to a pixel and therefore has known image and object coordinates. Array 836 in this disclosure is referred to as “array space”. The array space data has a one-to-one correspondence with the pixel space data.
Image processing including enhancement, restoration, and segmentation methods are referenced in many texts and is part of filtering program 835.
Using array 836 data, numerous analytical geometric and array type calculations can be performed by image analysis program 812. For example, the center point 838 of vertical boundary line segment 839 can be determined as well as the center point 842 of vertical boundary line segment 844 by counting the number of 1's in the vertical boundary line columns and dividing this number by 2, thereby determining the lateral center line location of the roadway mark path 16 in array and/or object space.
As another example, the bottom and top location of the vertical boundary line 839 column is (18,9) and (8,9) respectively. The length is computed by taking the difference between these array elements and equals 10 array 836 units. This difference can also be used to determine the width of the roadway mark in either image, array or object space using program 810. Image, array and object space distances and coordinates of lines may be determined from array 836. Additionally, other geometric features may be defined and their respective array and/or object space coordinates identified. For example, coordinates of line segment intersections can be determined.
Image analysis program 812 also determines both the beginning and ending lateral edge lines 42 and 46, and the center points 43 and 47 of the beginning and ending lateral edge lines 42 and 46, respectively, from a sequence of roadway mark segment 18 images and their corresponding array 836 data of the roadway mark segment 18. This data is useful for restriping over the currently visible pre-existing roadway mark without regard for the line strip pattern (i.e., mark segment 18 and gap segment 20 lengths). Additionally, the image analysis program 812 also determines the image, array, and object space coordinates of lines 42 and 46 and center points 43 and 47, and uses data from pixel-to-distance program 810 to determine object space coordinates and distances.
In addition, program 812 determines the pixel, array and object space coordinates of the imaged laser line pattern 106a from the corrected image and determines its corresponding array and object space coordinates. The results of image analysis program 812 are stored in data memory 806.
Image analysis program 812 also inputs pulse count data from drive shaft positional sensor 500 and can perform calculations using these and other data. For example, the image analysis program 812 can determine the coordinate differences between common features of successive images in array and/or object space, and determine the number of shaft positional sensor 500 pulses which occurred between the successive image common features.
Image analysis program 812 also determines the number of shaft positional sensor 500 pulses which is required to travel a given distance in array or object space. For example, if the desired distance to be travelled equals 50 inches (object space coordinates) and the pulse to distance ratio is 20 pulses per inch (derived from calibration), then the number of pulses required to travel 50 inches equals 1,000 pulses (50 inches×20 pulses per inch).
Image analysis program 812 also may determine the type of roadway mark pattern, i.e., if the imaged roadway mark segment 18 comprises a single or double line, a solid or skip line, or any combination and the line patterns using image, array, and/or object space calculations and conventional image and array processing algorithms. The type of line being imaged, its location and time of acquisition is stored in data memory 806.
Image analysis program 812 also determines the speed of vehicle 50 by determining the array 836 coordinate differences between features of successive images of the roadway mark segment 18 (for example the image of the beginning line 42) and converting this difference to object space distance using data from pixel-to-distance program 810, and then taking the difference in time between the successive images.
The time each image was acquired and time interval between images are determined by the timing of a trigger signal placed onto trigger input 728, or by other aforementioned deterministic-derived trigger signals previously mentioned (for example, timing signals derived from GPS receiver 154 or PTP unit 111) and are known by image analysis program 812. The speed of vehicle 50 is then determined knowing both the object distance travelled and the amount of time to travel this distance (distance/time). Data from GPS receiver 154 may also be used to determine distances and time intervals, and therefore the speed of vehicle 50.
Image analysis program 812 can also analyze array 836 data produced from a sequence of images and determine the gap and mark segment lengths and determine the skip line pattern (for example, a 15/40 pattern).
As previously noted, program 812 includes a machine learning network for classifying individual pixels (referred to semantic classification) of the roadway surface image.
An example of a machine learning network is illustrated in
The roadway mark image 14A is first input into the conventional convolutional block 14B which then flows to the depth-wise separable convolution block 14C, the output of which then subsequently flows to another depth-wise separable convolution block 14D.
The output from the convolution block 14D then splits between two branches. The upper branch comprises concatenated inverted residual bottleneck blocks 14E. The output data of the bottleneck blocks 14E then flows to the pyramid pooling block 14F. The output of the pyramid pooling block 14F then flows to the up-sample program block 14G, the output of which flows to the depth-wise convolution block 14H. The output from the convolution block 14H then flows to the conventional convolutional block 14I.
The lower branch comprises one conventional convolution block 14J.
The outputs of the conventional convolutional blocks 14I and 14J are then combined and flow to the depth-wise separable convolution block 14K. The output of the convolution block 14K then flows to the input of another depth-wise separable convolutional block 14L. The output of the convolutional 14L then flows to the input of the conventional convolutional block 14M. The output of the convolutional block 14M then flows to the input of the up-sample program block 14N. The output from the program block 14N then flows to the softmax program block 14O.
Software implementation of all block components 14A through 14O are known in the computer vision art.
An example of a Fast-SCNN architecture is reviewed in the article entitled “Fast-SCNN: Fast Semantic Segmentation Network” by Rudra P K Poudel, Stephen Liwicki and Roberto Cipolla, arXiv:1902.04502v1 (12 Feb. 2019), and is available in PyTorch, an open-source deep machine learning library offered by Facebook's AI Research Laboratory.
Mark path projection program 814 computes an equation which models the position of the roadway mark path 16 in array space 836 (and/or object space) based upon a sequence of array coordinates of individual center points 838 and 842 (or other common features) sequentially determined from a sequence of corrected images and image analysis program 812.
Other available array 836 element coordinates may be defined and used in these calculations. For example, the coordinates of corners may be used. In addition, mark path projection program 814 may also determine an equation which predicts the mark path 16 projection in array space (and/or object space using pixel-to-distance program 810).
For example, two coordinate pairs of similar features in array space may be used to develop a mathematical model (for example, a conventional y=mx+b linear equation) of the roadway mark path 16 in either array or object space, and three coordinate pairs may be used to develop a quadratic or other type of interpolated curvature model. This information is then used along with the image space-to-object space conversion values from pixel-to-distance program 810 to develop an array and/or object space projection model of the roadway mark path and is used to project the mark path location rearward of vehicle 50 especially under the area of carriage 80 including the paint guns 84 and 86 and the bead guns 88 and 90 dispensing area.
Machine vision and carriage control program 816 positions carriage 80 so that paint gun 84 and its associated bead gun 88 are placed over a pre-existing single roadway mark segment 18. It is assumed that paint gun 86 and its bead gun 90 have been laterally adjusted to accommodate a second roadway mark if gun 84 is properly aligned with its respective roadway mark segment 18. Any number of paint and/or bead guns may be accommodated. Machine vision and carriage control program 816 may use either the array and/or object space data 836 and roadway mark path 16 mathematical projection model representation acquired from the mark path projection program 814.
Machine vision and carriage control program 816 also computes the array and/or object space coordinates of the intersection point between the lateral object space projection line 81 equation and the roadway mark path 16 object space path projection equation. The projection line 81 equation is computed from object space calibration points (and may also be transformed into array space coordinates for an “extended” array space underneath the carriage) and is known.
The coordinates of the intersection point define the alignment location of paint gun 84 and its respective nozzle (and bead gun 88) to dispense roadway mark material directly over roadway mark segment 18.
The machine vision and carriage control program 816 also computes a difference between the lateral coordinates of the intersection point and the current lateral position of carriage 80. The current lateral position of carriage 80 is known from the linear position sensor 110, and therefore, the current position of paint guns 84 and 86, and their respective nozzles 88 and 90 (offset adjusted) are known. Alternately, the current carriage 80 lateral position may be computed by image analysis program 812 using coordinates of the corrected imaged laser line pattern 106a of laser line pattern 106.
This lateral difference is an error signal ϵ and, for proper alignment of the paint gun 88 over the roadway mark segment 18, should be equal to 0.
The machine vision and carriage control program 816 also includes a carriage position controller which, in response to the error signal ϵ, sends positional commands to the hydraulic system 401 (specifically to control valve 427 via the CANopen interface 429) to either extend, retract or idle the position of carriage 80 to force the error signal ϵ to 0.
It is therefore understood that the machine vision and carriage control program 816 can extend, retract, or idle carriage 80 to laterally align the paint gun 84 (and also the bead gun 88) with the roadway mark segment 18.
Referring to
The control system 1700 comprises a mark path projection system 1701 (which may comprise mark path projection program 814), a carriage control system 1720 (which may comprise machine vision and carriage control program 816), hydraulic system 401, the camera or imager 252, the image correction system 1725 (which may comprise image correction program 808), and the image analysis system 1730 (which may comprise image analysis program 812). The system 1720 further comprises a mark alignment calculator 1703, the comparator 1705, and the carriage position controller 1710. The systems 1701, 1720, 1725, and 1730 may be implemented in software, hardware (such as an FPGA), or a combination of software and hardware.
Mark path projection system 1701 inputs data from image analysis system 1730 via a line 1740 and creates a roadway mark path 16 mathematical projection model (equation).
The roadway mark path 16 model is then used by mark alignment calculator 1703 to calculate the intersection point between the calculated projection line 81 equation and the roadway mark segment 18 path projection equation. The calculated intersection point represents the lateral position of the actual roadway mark segment 18 along the line 81 as it passes under carriage 80. The intersection point calculation may be accomplished in array and/or object space coordinates.
The calculated intersection point is the desired lateral position of paint gun 84 and its respective nozzle to dispense the roadway mark material (paint) directly over and onto the pre-existing roadway mark segment 18. The desired lateral position array and/or object space coordinate data are then input into the positive (+) input of comparator 1705.
Comparator 1705 takes the difference between the lateral distance of the intersection point along line 81 and the actual lateral position of paint gun 84 determined by the linear position sensor 110. This difference is an error signal E which is placed onto line 1707 and which flows to the input of the carriage position controller 1710. The error signal E may be expressed in array and/or object space coordinates.
In response to the error signal E placed on line 1707, the carriage position controller 1710 sends positional commands to the hydraulic system 401 (specifically to control valve 427 via the CANopen interface 429) to either extend, retract or idle the position of carriage 80.
In response to the positional commands received from carriage position controller 1710, hydraulic system 401 via valve 427 either extends or retracts the piston 447 thus changing the lateral position of the paint gun 84 (and also bead gun 88) assuming a non-zero error signal 1707. If the error signal 1707 equals zero, the hydraulic system 401 maintains the current carriage 80 lateral position and therefore the current lateral position of paint gun 84 (and also bead gun 88). The carriage position controller 1710 provides smooth and coordinated movements with a carriage acceleration and velocity profile consistent with roadway marking systems.
The mark path projection system 1701, the mark alignment calculator 1703, the comparator 1705, the carriage position controller 1710, and the hydraulic system 401, and the linear position sensor 110 all cooperate to move the carriage 80 is a direction to minimize the error signal E.
As the carriage 80 is being moved by the hydraulic system 410, the imager 252 continually images the position of the roadway mark 18 and the mark path projection system 1701 updates the current position of the roadway mark 18, recalculating the roadway mark path 16 mathematical projection model.
It is therefore understood that the location of carriage 80 is automatically adjusted to correctly position paint gun 84 and its respective nozzle over a projected roadway mark segment 18 using machine learning network technology.
Referring to
The machine-learning-network based carriage control system 1750 is similar to the control system 1700 except that the current lateral position of carriage 80 is determined from the coordinates of the image 106a of the laser line 106 (offset corrected). The coordinates of the imaged laser line 106a may be expressed in array or object space coordinates.
The imager 252 images both the roadway mark 18 and the laser line 106 combination and outputs this image to the image correction system 1725. The output from the image correction system 1725 then flows to the input of the image analyses system 1730.
The image analysis program 1730 calculates both the coordinates of the roadway mark 18 and the imaged laser line pattern 106a in both array and object space coordinates. The image analysis program 1730 then outputs the laser line pattern 106a coordinates to the (−) input of the comparator 1705 via line 1735, and outputs the roadway mark 18 coordinates to the mark path projection system 1701 via line 1740.
Changing the lateral position of carriage 80 laterally moves imaged laser line pattern 106a, and the carriage control system 1720 moves carriage 80 in a lateral direction which minimizes error signal 1707 thereby aligning paint gun 84 and bead gun 88 with roadway mark segment 18 as in a conventional classical (servo) feedback system.
It is therefore understood that the location of carriage 80 is automatically adjusted to correctly laterally position paint gun 84 and bead gun 88 (along with their respective nozzles) over a roadway mark segment 18 using machine learning network technology. It is now necessary to determine when to turn-on and turn-off the paint gun 84 and the associated bead gun 88 to correctly and accurately duplicate a pre-existing roadway mark segment 18. It is understood that the control of the paint gun 86 and bead gun 90 is similarly controlled by system 700.
Dynamic positional calibration program 818 dynamically calibrates the pulse-to-distance ratio of drive shaft positional sensor 500 by computing the pixel difference in array space between common features of roadway mark images, such as center point 838 in
For example, in
This technique does not rely upon tire diameter or pressure and is therefore more accurate and dynamic than conventional methods, i.e., this technique auto-calibrates with every image taken which has an identifiable feature for each image. A new current distance-per-pulse ratio is therefore calculated continuously with each image and is not a fixed value as currently assumed in the industry. Alternately, the number of array 836 elements per pulse may also be determined (6 array elements/50 pulses=0.12 array elements per pulse).
Dispensing control program 820 controls pressurized control system 460 (via an I/O board in computer 703 as previously disclosed) and determines which solenoid valves to activate and the time duration. Turn-on and turn-off delays of the valves are also determined.
Referring now to
From the image of area 70, the object space location of roadway mark segment 18 beginning line 42 is determined relative to the origin 1002 of the array and/or object space coordinate system (programs 810 and 812 perform this image space-to-object space transformation), and in particular the longitudinal distance 1004 from lateral edge line 42 to origin 1002 is determined. The location of the ending edge line 46 is similarly determined from the array and/or object space coordinate system.
The longitudinal distance 1006 from gun 84 to origin 1002 has been previously determined via a calibration procedure, and the distance 1008 has been determined by using a ruler or other calibration methods. Therefore, distances 1010 and 1012 are simply determined by adding distances 1004 and 1006 and adding distances 1004, 1006, and 1008, respectively.
The moment of when to turn-on paint gun 84 is when edge line 42 in under paint gun 84 or, equivalently, when line 42 has travelled a total distance 1010. The number of pulses required to travel the distance 1010 is determined by the distance-to-pulse ratio previously determined in dynamic positional calibration program 818. When the number of pulses required to travel the distance 1010 has been accumulated, the dispensing control program 820 generates a begin dispensing signal to start dispensing roadway mark material using air control system 160
Similarly, distance 1012 is calculated by counting pulses 534 equivalent to distance 1012. Travelled distance can also be calculated using the GPS receiver 154 and IMU 155 combination. Dispensing control program 820 also takes into consideration the turn-on and turn-off times of the respective guns. Equivalent calculations may be also performed in array 836 space.
Dispensing continues until the number of pulses 534 equals the equivalent distance of roadway mark segment length 24 which has been previously input into computer 703 via keyboard 706 by the operator, or when the system detects the lateral edge line 46. Also note that dispensing does not occur for the next mark segment 32 until an accumulated pulse count equal to the distance of roadway mark gap segment length 26 has been obtained, or when the system detects the beginning lateral edge line 42 of the next roadway mark segment. Because the distance-to-pulse ratio is continuously updated and dynamically calculated, accurate maintenance striping of the roadway mark elements occurs without the need for additional carriage operators to force a lead or lag time adjustment during the dispensing cycle.
Although the above discussion refers to a single skip line roadway mark segment 18, similar procedures can be implemented to apply roadway mark material to a double skip line mark, or to single or double solid line mark(s) or combinations.
For clarity purposes in understanding the operation of the preferred embodiment of the invention, only two successive images of roadway mark segment 18 will be used.
In operation and referring additionally to
In step 1100, the operator positions vehicle 50 along a desired roadway mark path 16 having a roadway mark which needs to be restriped with location assistance being supplied by GPS receiver 154 and IMU 155 and LCD 704. It is assumed that all calibration and offset data have been previously obtained and are stored in data memory 806. The driver then inputs the desired line stripe pattern (single or double, solid or skip line(s), for example 15/40) and the type of roadway mark material (paint or paint and bead) to be dispensed using keyboard 706 and depresses the start button on keyboard 706. The desired line stripe pattern is entered into the image analysis system 1730.
Alternately, the driver may decide to enable the system 700 to determine the line stripe pattern by determining the length 24 of mark segment 18 and length 26 of gap segment 20 using the image analysis system 1730. Yet further, the driver may decide to enable the system 700 to restripe over the currently visible pre-existing roadway mark without regard for the mark segment 18 and gap segment 20 lengths using image analysis system 1730.
The driver then proceeds to drive vehicle 50 along roadway mark path 16 at a vehicle speed consistent with the type of roadway mark material being used for the restriping process. Continuously updated GPS positional data may be displayed on LCD display 704 (as is currently available in automobiles) to assist the driver in positioning vehicle 50 at the correct starting location and along the roadway mark path 16. Program flow then continues to step 1102.
In step 1102 and in response to the “Start Button” being depressed on the keyboard 706, the imager 252 of the system 700 (and system 750 if vehicle 50 is so equipped) acquires a first time stamped raw image data of the beginning of roadway mark segment 18. GPS positional data from GPS receiver 154 is also acquired, and along with the raw image data, is stored in memory 806. Program flow then continues to step 1104.
In step 1104, the first time stamped raw image data are undistorted by image correction system 1725 and the first time stamped undistorted image of roadway mark segment 18 is stored in data memory 806 along with the time stamp and corresponding GPS positional data. Program flow then continues to step 1106.
In step 1106, the array 836 coordinates of a first center point 838 (see
In step 1108, system 700 (and system 750 if vehicle 50 is so equipped) acquires a second timed stamped raw image of roadway mark segment 18 and GPS positional data. This second image occurs after, and is displaced from, the first image because vehicle 50 is moving along the roadway mark path 16 in direction 28. Program flow then continues to step 1110.
In step 1110, the second time stamped raw image data are undistorted by image correction system 1725 and the raw and undistorted images of roadway mark segment 18 are stored in data memory 806 along with the time stamp and GPS positional data. Program flow then continues to step 1112.
In step 1112, the array 836 coordinates of a second displaced center point 838 of the second image of roadway mark segment 18 are determined by image analysis system 1730. Program flow then continues to step 1114. Note that other common points or features in consecutive images may be used in place of the center points 838. Program flow then continues to step 1114.
In step 1114, a roadway mark path 16 projection equation in array space is calculated using the first and displaced second image center points 838 (in this case the equation will be a line) using mark path projection system 1701. Program flow continues to step 1116.
In step 1116 and based upon the mark path projection equation derived from mark path projection system 1701 and the previously stored lateral projection line 81 equation, an intersection point is determined between these two equations in array (and/or object) space using mark alignment calculator 1703 of the carriage control system 1720.
The comparator 1705 inputs the intersection point coordinates and the linear position sensor 110 data and then takes the difference between the intersection point (the desired carriage 80 position) and the linear position sensor 110 data (the actual carriage 80 position) and forms an error signal 1707. The error signal 1707 signal then flows to the input of the carriage position controller 1710.
In response to the error signal 1707 (ε), the carriage position controller 1710 sends positioning commands to the hydraulic system 401 to either extend, retract, or idle carriage 80 and to align paint gun 84 (and therefore also bead gun 88) directly over roadway mark segment 18. Program flow continues to step 1118.
In step 1118, the array 836 coordinates of the first center point 838 are subtracted from the array 836 coordinates of the second center point 838 of the roadway mark segment 18 and the number of system 500 generated pulses occurring for this difference is determined by image analysis system 1730. Program flow continues to step 1120.
In step 1120, the array (and/or object) space distance from the second center point 838 to the intersection point between the mark path projection equation and the previously stored lateral projection line 81 is calculated by image analysis system 1730. Program flow continues to step 1122.
In step 1122, the number of generated distance pulses required to cover the distance from the last imaged center point 838 to the intersection point along the mark path projection line of roadway mark path 16 is determined by image analysis system 1730. Program flow continues to step 1124.
In step 1124, when the number of drive shaft positional sensor 500 generated pulses has occurred as determined in step 1122, dispensing control program 820 controls the pressurized air flow via system 460 to gun 84 (and bead gun 88 if required, as previously input by the operator in step 1100). In response to the pressurized air, gun 84 (and bead gun 88 if required) begins dispensing roadway mark material onto, and in alignment with, roadway mark segment 18 until the number of system 500 pulses equals the desired mark segment length in array and/or object space as previously input by the operator, or determined by analyzing a sequence of roadway images using the image analysis program 812. Program flow then continues to step 1126.
In step 1126, after the number of system 500 pulses equals the required distance of mark segment 18 having length 24, paint gun 84 is turned off (and also its associated bead gun 88 if previously enabled) ceasing material dispensing for a number of drive shaft positional sensor 500 pulses equal to the length of the gap segment 20. Another dispensing cycle begins and continues until the desired entire length of center skip line 12 has been restriped.
Machine vision and machine learning based control system 1700 continually updates the lateral position of carriage 80 and therefore the lateral position of the paint gun 84 (and its respective nozzle) and the bead gun 88 to continuously dispense roadway mark material directly over and onto a solid roadway mark segment.
For the second embodiment of the invention, step 1116 is modified by having the comparator 1705 input the coordinates of the imaged laser line 106a determined by the image analysis system 1730 instead of using the linear position sensor 110 data.
It is therefore understood that a single line, double line or skip-line patterns or combinations thereof such may be restriped according to the teachings of this invention.
Although illustrated and described above with reference to certain specific embodiments and examples, the present invention is nevertheless not intended to be limited to the details shown. Rather, various modifications may be made in the details within the scope and range of equivalents of the claims and without departing from the spirit of the invention. It is expressly intended, for example, that all ranges broadly recited in this document include within their scope all narrower ranges which fall within the broader ranges.
This application claims the benefit of U.S. Provisional Application No. 63/010,876, titled, “Machine Learning Based Carriage Control Apparatus For Maintenance Striping,” filed Apr. 16, 2020, the content of which is incorporated herein by reference in its entirety.
| Number | Name | Date | Kind |
|---|---|---|---|
| 3101175 | Brown | Aug 1963 | A |
| 3229660 | McLucas et al. | Jan 1966 | A |
| 5054959 | Wilson | Oct 1991 | A |
| 5169262 | Wilson et al. | Dec 1992 | A |
| 5203923 | Hartman | Apr 1993 | A |
| 5456548 | Smyrk | Oct 1995 | A |
| 6597143 | Song | Jul 2003 | B2 |
| 20210200624 | Feekes, Jr. | Jul 2021 | A1 |
| Entry |
|---|
| Dosovitskiy et al., “CARLA: An Open Urban 1st Driving Simulator” Conference on Robot Learning, Mountain View, California (2017). |
| Poudel et al., “Fast-SCNN: Fast Semantic Segmentation Network” https://arxiv.org/abs/1902.04502 (2019). |
| Number | Date | Country | |
|---|---|---|---|
| 20210326606 A1 | Oct 2021 | US |
| Number | Date | Country | |
|---|---|---|---|
| 63010876 | Apr 2020 | US |
