Patent Application
20200310621
This disclosure relates generally to detecting touch events in a touch-sensitive device, and in particular to classifying wanted and unwanted touches.
Touch-sensitive displays for interacting with computing devices are becoming more common. A number of different technologies exist for implementing touch-sensitive displays and other touch-sensitive devices. Examples of these techniques include, for example, resistive touch screens, surface acoustic wave touch screens, capacitive touch screens and certain types of optical touch screens.
However, many of these approaches currently suffer from drawbacks. For example, some technologies may function well for small sized displays, as used in many modern mobile phones, but do not scale well to larger screen sizes as in displays used with laptop or even desktop computers. For technologies that require a specially processed surface or the use of special elements in the surface, increasing the screen size by a linear factor of N means that the special processing must be scaled to handle the N2 larger area of the screen or that N2 times as many special elements are required. This can result in unacceptably low yields or prohibitively high costs.
Another drawback for some technologies is their inability or difficulty in handling multitouch events. A multitouch event occurs when multiple touch events occur simultaneously. This can introduce ambiguities in the raw detected signals, which then must be resolved. Furthermore, there are limits on the time available for resolving these ambiguities. If the approach adopted is too slow, then the technology will not be able to deliver the touch sampling rate required by the system. If the approach adopted is too computationally intensive, then this will drive up the cost and power consumption of the technology.
Embodiments relate to classifying touch events on or near a touch surface as wanted or unwanted touch events. An example touch-sensitive device is an optical touch-sensitive device that is able to determine the locations of multiple simultaneous touch events. The optical touch-sensitive device may include multiple emitters and detectors. Each emitter produces optical beams which are received by the detectors. The optical beams preferably are multiplexed in a manner so that many optical beams can be received by a detector simultaneously. Touch events disturb the optical beams.
Embodiments relate to a method for detecting touch events on or near a surface. The surface has one or more emitters and one or more detectors. The emitters produce optical beams that propagate along the surface and are received by the detectors. Touch events disturb the optical beams. One or more beam responses are measured. A location of a first touch event and a location of an additional touch event are estimated based on the one or more beam responses. A shared beam of the one or more beam responses is identified. The shared beam is associated with the first touch event and the additional touch event. The one or more beam responses are compensated based on the identification of the shared beam. An updated location of the first touch event is determined based on the compensated one or more beam responses.
In some embodiments, compensating the one or more beam responses based on the identification of the shared beam includes removing the beam response of the shared beam from the one or more beam responses.
In some embodiments, compensating the one or more beam responses based on the identification of the shared beam includes removing a portion of a beam response of the shared beam from the one or more beam responses. In some embodiments, compensating the one or more beam responses based on the identification of the shared beam further includes determining a contribution of the additional touch event to the beam response of the shared beam, where the removed portion of the beam response of the shared beam is the contribution of the additional touch event. In some embodiments, locations of touch events in previous frames are referenced. The location of the additional touch event is determined to be within a threshold distance of a location of a touch event in a previous frame. The additional touch event is classified as a virtual touch caused by contamination on the screen.
In some embodiments, estimating a location of a first touch event and a location of an additional touch event based on the one or more beam responses includes determining an activity map based on the one or more beam responses. The activity map represents touch events on or near the surface. Additionally, the estimated location of the first touch event and the estimated location of the additional touch event is determined based on the activity map. In some embodiments, updating the location of the first touch event based on the compensated one or more beam response includes re-determining the activity map based on the compensated one or more beam responses. Additionally, the updated location of the first touch event is determined based on the re-determined activity map.
In some embodiments, the one or more beam responses are measured for a current frame and are measured relative to a baseline beam response. The baseline beam response is based on one or more beam responses measured for a past frame.
Embodiments of the present invention will now be described, by way of example, with reference to the accompanying drawings, in which:
A. Device Overview
The emitter/detector drive circuits 120 serve as an interface between the controller 110 and the emitters Ej and detectors Dk. The emitters produce optical “beams” which are received by the detectors. Preferably, the light produced by one emitter is received by more than one detector, and each detector receives light from more than one emitter. For convenience, “beam” will refer to the light from one emitter to one detector, even though it may be part of a large fan of light that goes to many detectors rather than a separate beam. The beam from emitter Ej to detector Dk will be referred to as beam jk.
The emitters and detectors may be interleaved around the periphery of the sensitive surface. In other embodiments, the number of emitters and detectors are different and are distributed around the periphery in any defined order. The emitters and detectors may be regularly or irregularly spaced. In some cases, the emitters and/or detectors may be located on less than all of the sides (e.g., one side). In some embodiments, the emitters and/or detectors are not located around the periphery (e.g., beams are directed to/from the active touch area 131 by optical beam couplers). Reflectors may also be positioned around the periphery to reflect optical beams, causing the path from the emitter to the detector to pass across the surface more than once.
One advantage of an optical approach as shown in
B. Process Overview
The transmission coefficient Tjk is the transmittance of the optical beam from emitter j to detector k, compared to what would have been transmitted if there was no touch event interacting with the optical beam. In the following examples, we will use a scale of 0 (fully blocked beam) to 1 (fully transmitted beam). Thus, a beam jk that is undisturbed by a touch event has Tjk=1. A beam jk that is fully blocked by a touch event has a Tjk=0. A beam jk that is partially blocked or attenuated by a touch event has 0<Tjk<1. It is possible for Tjk>1, for example depending on the nature of the touch interaction or in cases where light is deflected or scattered to detectors k that it normally would not reach.
The use of this specific measure is purely an example. Other measures can be used. In particular, since we are most interested in interrupted beams, an inverse measure such as (1−Tjk) may be used since it is normally 0. Other examples include measures of absorption, attenuation, reflection, or scattering. In addition, although
Returning to
For example, the physical phase 210 produces transmission coefficients Tjk. Many different physical designs for the touch-sensitive surface assembly 130 are possible, and different design tradeoffs will be considered depending on the end application. For example, the emitters and detectors may be narrower or wider, narrower angle or wider angle, various wavelengths, various powers, coherent or not, etc. As another example, different types of multiplexing may be used to allow beams from multiple emitters to be received by each detector. Several of these physical setups and manners of operation are described below, primarily in Section II.
The interior of block 210 shows one possible implementation of process 210. In this example, emitters transmit 212 beams to multiple detectors. Some of the beams travelling across the touch-sensitive surface are disturbed by touch events. The detectors receive 214 the beams from the emitters in a multiplexed optical form. The received beams are de-multiplexed 216 to distinguish individual beams jk from each other. Transmission coefficients Tjk for each individual beam jk are then determined 218.
The processing phase 220 can also be implemented in many different ways. Candidate touch points, line imaging, location interpolation, touch event templates, and multi-pass approaches are all examples of techniques that may be used as part of the processing phase 220. Several of these are described below, primarily in Section III.
The touch-sensitive device 100 may be implemented in a number of different ways. The following are some examples of design variations.
A. Electronics
With respect to electronic aspects, note that
For example, the controller 110 and touch event processor 140 may be implemented as hardware, software or a combination of the two. They may also be implemented together (e.g., as a SoC with code running on a processor in the SoC) or separately (e.g., the controller as part of an ASIC, and the touch event processor as software running on a separate processor chip that communicates with the ASIC). Example implementations include dedicated hardware (e.g., ASIC or programmed field programmable gate array (FPGA)), and microprocessor or microcontroller (either embedded or standalone) running software code (including firmware). Software implementations can be modified after manufacturing by updating the software.
The emitter/detector drive circuits 120 serve as an interface between the controller 110 and the emitters and detectors. In one implementation, the interface to the controller 110 is at least partly digital in nature. With respect to emitters, the controller 110 may send commands controlling the operation of the emitters. These commands may be instructions, for example a sequence of bits which mean to take certain actions: start/stop transmission of beams, change to a certain pattern or sequence of beams, adjust power, power up/power down circuits. They may also be simpler signals, for example a “beam enable signal,” where the emitters transmit beams when the beam enable signal is high and do not transmit when the beam enable signal is low.
The circuits 120 convert the received instructions into physical signals that drive the emitters. For example, circuit 120 might include some digital logic coupled to digital to analog converters, in order to convert received digital instructions into drive currents for the emitters. The circuit 120 might also include other circuitry used to operate the emitters: modulators to impress electrical modulations onto the optical beams (or onto the electrical signals driving the emitters), control loops and analog feedback from the emitters, for example. The emitters may also send information to the controller, for example providing signals that report on their current status.
With respect to the detectors, the controller 110 may also send commands controlling the operation of the detectors, and the detectors may return signals to the controller. The detectors also transmit information about the beams received by the detectors. For example, the circuits 120 may receive raw or amplified analog signals from the detectors. The circuits then may condition these signals (e.g., noise suppression), convert them from analog to digital form, and perhaps also apply some digital processing (e.g., demodulation).
B. Touch Interactions
For example,
In
In
The touch mechanism may also enhance transmission, instead of or in addition to reducing transmission. For example, the touch interaction in
For simplicity, in the remainder of this description, the touch mechanism will be assumed to be primarily of a blocking nature, meaning that a beam from an emitter to a detector will be partially or fully blocked by an intervening touch event. This is not required, but it is convenient to illustrate various concepts.
For convenience, the touch interaction mechanism may sometimes be classified as either binary or analog. A binary interaction is one that basically has two possible responses as a function of the touch. Examples includes non-blocking and fully blocking, or non-blocking and 10%+ attenuation, or not frustrated and frustrated TIR. An analog interaction is one that has a “grayscale” response to the touch: non-blocking passing through gradations of partially blocking to blocking. Whether the touch interaction mechanism is binary or analog depends in part on the nature of the interaction between the touch and the beam. It does not depend on the lateral width of the beam (which can also be manipulated to obtain a binary or analog attenuation, as described below), although it might depend on the vertical size of the beam.
C. Emitters, Detectors, and Couplers
Each emitter transmits light to a number of detectors. Usually, each emitter outputs light to more than one detector simultaneously. Similarly, each detector receives light from a number of different emitters. The optical beams may be visible, infrared, and/or ultraviolet light. The term “light” is meant to include all of these wavelengths and terms such as “optical” are to be interpreted accordingly.
Examples of the optical sources for the emitters include light emitting diodes (LEDs) and semiconductor lasers. IR sources can also be used. Modulation of optical beams can be achieved by directly modulating the optical source or by using an external modulator, for example a liquid crystal modulator or a deflected mirror modulator. Examples of sensor elements for the detector include charge coupled devices, photodiodes, photoresistors, phototransistors, and nonlinear all-optical detectors. Typically, the detectors output an electrical signal that is a function of the intensity of the received optical beam.
The emitters and detectors may also include optics and/or electronics in addition to the main optical source and sensor element. For example, optics can be used to couple between the emitter/detector and the desired beam path. Optics can also reshape or otherwise condition the beam produced by the emitter or accepted by the detector. These optics may include lenses, Fresnel lenses, mirrors, filters, non-imaging optics, and other optical components.
In this disclosure, the optical paths will be shown unfolded for clarity. Thus, sources, optical beams, and sensors will be shown as lying in one plane. In actual implementations, the sources and sensors typically will not lie in the same plane as the optical beams. Various coupling approaches can be used. A planar waveguide or optical fiber may be used to couple light to/from the actual beam path. Free space coupling (e.g., lenses and mirrors) may also be used. A combination may also be used, for example waveguided along one dimension and free space along the other dimension. Various coupler designs are described in U.S. Application Ser. No. 61/510,989 “Optical Coupler” filed on Jul. 22, 2011, which is incorporated by reference in its entirety herein.
D. Optical Beam Paths
Another aspect of a touch-sensitive system is the shape and location of the optical beams and beam paths. In
Note that every emitter Ej may not produce beams for every detector Dk. In
The footprints of individual beams from an emitter and the coverage area of all beams from an emitter can be described using different quantities. Spatial extent (i.e., width), angular extent (i.e., radiant angle for emitters, acceptance angle for detectors) and footprint shape are quantities that can be used to describe individual beam paths as well as an individual emitter's coverage area.
An individual beam path from one emitter Ej to one detector Dk can be described by the emitter Ej's width, the detector Dk's width and/or the angles and shape defining the beam path between the two.
These individual beam paths can be aggregated over all detectors for one emitter Ej to produce the coverage area for emitter Ej. Emitter Ej's coverage area can be described by the emitter Ej's width, the aggregate width of the relevant detectors Dk and/or the angles and shape defining the aggregate of the beam paths from emitter Ej. Note that the individual footprints may overlap (see
The coverage areas for individual emitters can be aggregated over all emitters to obtain the overall coverage for the system. In this case, the shape of the overall coverage area is not so interesting because it should cover the entirety of the active area 131. However, not all points within the active area 131 will be covered equally. Some points may be traversed by many beam paths while other points traversed by far fewer. The distribution of beam paths over the active area 131 may be characterized by calculating how many beam paths traverse different (x,y) points within the active area. The orientation of beam paths is another aspect of the distribution. An (x,y) point that is derived from three beam paths that are all running roughly in the same direction usually will be a weaker distribution than a point that is traversed by three beam paths that all run at 60 degree angles to each other.
The discussion above for emitters also holds for detectors. The diagrams constructed for emitters in
A detector Dk's coverage area is then the aggregate of all footprints for beams received by a detector Dk. The aggregate of all detector coverage areas gives the overall system coverage.
E. Active Area Coverage
The coverage of the active area 131 depends on the shapes of the beam paths, but also depends on the arrangement of emitters and detectors. In most applications, the active area is rectangular in shape, and the emitters and detectors are located along at least a portion of the periphery of the rectangle.
In a preferred approach, rather than having only emitters along certain edges and only detectors along the other edges, emitters and detectors are interleaved along the edges.
F. Multiplexing
Since multiple emitters transmit multiple optical beams to multiple detectors, and since the behavior of individual beams is generally desired, a multiplexing/demultiplexing scheme is used. For example, each detector typically outputs a single electrical signal indicative of the intensity of the incident light, regardless of whether that light is from one optical beam produced by one emitter or from many optical beams produced by many emitters. However, the transmittance Tjk is a characteristic of an individual optical beam jk.
Different types of multiplexing can be used. Depending upon the multiplexing scheme used, the transmission characteristics of beams, including their content and when they are transmitted, may vary. Consequently, the choice of multiplexing scheme may affect both the physical construction of the optical touch-sensitive device as well as its operation.
One approach is based on code division multiplexing. In this approach, the optical beams produced by each emitter are encoded using different codes. A detector receives an optical signal which is the combination of optical beams from different emitters, but the received beam can be separated into its components based on the codes. This is described in further detail in U.S. Pat. No. 8,227,742 “Optical Control System With Modulated Emitters,” which is incorporated by reference herein.
Another similar approach is frequency division multiplexing. In this approach, rather than modulated by different codes, the optical beams from different emitters are modulated by different frequencies. The frequencies are low enough that the different components in the detected optical beam can be recovered by electronic filtering or other electronic or software means.
Time division multiplexing can also be used. In this approach, different emitters transmit beams at different times. The optical beams and transmission coefficients Tjk are identified based on timing. If only time multiplexing is used, the controller must cycle through the emitters quickly enough to meet the required touch sampling rate.
Other multiplexing techniques commonly used with optical systems include wavelength division multiplexing, polarization multiplexing, spatial multiplexing and angle multiplexing. Electronic modulation schemes, such as PSK, QAM and OFDM, may also be possibly applied to distinguish different beams.
Several multiplexing techniques may be used together. For example, time division multiplexing and code division multiplexing could be combined. Rather than code division multiplexing 128 emitters or time division multiplexing 128 emitters, the emitters might be broken down into 8 groups of 16. The 8 groups are time division multiplexed so that only 16 emitters are operating at any one time, and those 16 emitters are code division multiplexed. This might be advantageous, for example, to minimize the number of emitters active at any given point in time to reduce the power requirements of the device.
In the processing phase 220 of
A. Candidate Touch Points
One approach to determine the location of touch points is based on identifying beams that have been affected by a touch event (based on the transmission coefficients Tjk) and then identifying intersections of these interrupted beams as candidate touch points. The list of candidate touch points can be refined by considering other beams that are in proximity to the candidate touch points or by considering other candidate touch points. This approach is described in further detail in U.S. Pat. No. 8,350,831, “Method and Apparatus for Detecting a Multitouch Event in an Optical Touch-Sensitive Device,” which is incorporated herein by reference.
B. Line Imaging
This technique is based on the concept that the set of beams received by a detector form a line image of the touch points, where the viewpoint is the detector's location. The detector functions as a one-dimensional camera that is looking at the collection of emitters. Due to reciprocity, the same is also true for emitters. The set of beams transmitted by an emitter form a line image of the touch points, where the viewpoint is the emitter's location.
The example in
The touch point 910 casts a “shadow” in each of the lines images 1021-1023. One approach is based on finding the edges of the shadow in the line image and using the pixel values within the shadow to estimate the center of the shadow. A line can then be drawn from a location representing the beam terminal to the center of the shadow. The touch point is assumed to lie along this line somewhere. That is, the line is a candidate line for positions of the touch point.
Each line image shown in
C. Location Interpolation
Applications typically will require a certain level of accuracy in locating touch points. One approach to increase accuracy is to increase the density of emitters, detectors and beam paths so that a small change in the location of the touch point will interrupt different beams.
Another approach is to interpolate between beams. In the line images of
The interpolation accuracy can be enhanced by accounting for any uneven distribution of light across the beams a2 and b1. For example, if the beam cross section is Gaussian, this can be taken into account when making the interpolation. In another variation, if the wide emitters and detectors are themselves composed of several emitting or detecting units, these can be decomposed into the individual elements to determine more accurately the touch location. This may be done as a secondary pass, having first determined that there is touch activity in a given location with a first pass. A wide emitter can be approximated by driving several adjacent emitters simultaneously. A wide detector can be approximated by combining the outputs of several detectors to form a single signal.
D. Touch Event Templates
If the locations and shapes of the beam paths are known, which is typically the case for systems with fixed emitters, detectors, and optics, it is possible to predict in advance the transmission coefficients for a given touch event. Templates can be generated a priori for expected touch events. The determination of touch events then becomes a template matching problem.
If a brute force approach is used, then one template can be generated for each possible touch event. However, this can result in a large number of templates. For example, assume that one class of touch events is modeled as oval contact areas and assume that the beams are pencil beams that are either fully blocked or fully unblocked. This class of touch events can be parameterized as a function of five dimensions: length of major axis, length of minor axis, orientation of major axis, x location within the active area and y location within the active area. A brute force exhaustive set of templates covering this class of touch events must span these five dimensions. In addition, the template itself may have a large number of elements. Thus, it is desirable to simplify the set of templates.
Note that a series of templates could be defined for contact area 1210, increasing in the number of beams contained in the template: a 2-beam template, a 4-beam template, etc. In one embodiment, the beams that are interrupted by contact area 1210 are ordered sequentially from 1 to N. An n-beam template can then be constructed by selecting the first n beams in the order. Generally speaking, beams that are spatially or angularly diverse tend to yield better templates. That is, a template with three beam paths running at 60 degrees to each other and not intersecting at a common point tends to produce a more robust template than one based on three largely parallel beams which are in close proximity to each other. In addition, more beams tends to increase the effective signal-to-noise ratio of the template matching, particularly if the beams are from different emitters and detectors.
The template in
Other templates will be apparent and templates can be processed in a number of ways. In a straightforward approach, the disturbances for the beams in a template are simply summed or averaged. This can increase the overall SNR for such a measurement, because each beam adds additional signal while the noise from each beam is presumably independent. In another approach, the sum or other combination could be a weighted process, where not all beams in the template are given equal weight. For example, the beams which pass close to the center of the touch event being modeled could be weighted more heavily than those that are further away. Alternately, the angular diversity of beams in the template could also be expressed by weighting. Angular diverse beams are more heavily weighted than beams that are not as diverse.
In a case where there is a series of N beams, the analysis can begin with a relatively small number of beams. Additional beams can be added to the processing as needed until a certain confidence level (or SNR) is reached. The selection of which beams should be added next could proceed according to a predetermined schedule. Alternately, it could proceed depending on the processing results up to that time. For example, if beams with a certain orientation are giving low confidence results, more beams along that orientation may be added (at the expense of beams along other orientations) in order to increase the overall confidence.
The data records for templates can also include additional details about the template. This information may include, for example, location of the contact area, size and shape of the contact area and the type of touch event being modeled (e.g., fingertip, stylus, etc.).
In addition to intelligent design and selection of templates, symmetries can also be used to reduce the number of templates and/or computational load. Many applications use a rectangular active area with emitters and detectors placed symmetrically with respect to x and y axes. In that case, quadrant symmetry can be used to achieve a factor of four reduction. Templates created for one quadrant can be extended to the other three quadrants by taking advantage of the symmetry. Alternately, data for possible touch points in the other three quadrants can be transformed and then matched against templates from a single quadrant. If the active area is square, then there may be eight-fold symmetry.
Other types of redundancies, such as shift-invariance, can also reduce the number of templates and/or computational load. The template model of
In addition, the order of processing templates can also be used to reduce the computational load. There can be substantial similarities between the templates for touches which are nearby. They may have many beams in common, for example. This can be taken advantage of by advancing through the templates in an order that allows one to take advantage of the processing of the previous templates.
E. Multi-Pass Processing
Referring to
The first stage 1310 is a coarse pass that relies on a fast binary template matching, as described with respect to
Some simple clean-up 1316 is performed to refine this list. For example, it may be simple to eliminate redundant candidate touch points or to combine candidate touch points that are close or similar to each other. For example, the binary transmittances T′jk might match the template for a 5 mm diameter touch at location (x,y), a 7 mm diameter touch at (x,y) and a 9 mm diameter touch at (x,y). These may be consolidated into a single candidate touch point at location (x,y).
Stage 1320 is used to eliminate false positives, using a more refined approach. For each candidate touch point, neighboring beams may be used to validate or eliminate the candidate as an actual touch point. The techniques described in U.S. Pat. No. 8,350,831 may be used for this purpose. This stage may also use the analog values Tjk, in addition to accounting for the actual width of the optical beams. The output of stage 1320 is a list of confirmed touch points.
The final stage 1330 refines the location of each touch point. For example, the interpolation techniques described previously can be used to determine the locations with better accuracy. Since the approximate location is already known, stage 1330 may work with a much smaller number of beams (i.e., those in the local vicinity) but might apply more intensive computations to that data. The end result is a determination of the touch locations.
Other techniques may also be used for multi-pass processing. For example, line images or touch event models may also be used. Alternatively, the same technique may be used more than once or in an iterative fashion. For example, low resolution templates may be used first to determine a set of candidate touch locations, and then higher resolution templates or touch event models may be used to more precisely determine the precise location and shape of the touch.
F. Beam Weighting
In processing the transmission coefficients, it is common to weight or to prioritize the transmission coefficients. Weighting effectively means that some beams are more important than others. Weightings may be determined during processing as needed, or they may be predetermined and retrieved from lookup tables or lists.
One factor for weighting beams is angular diversity. Usually, angularly diverse beams are given a higher weight than beams with comparatively less angular diversity. Given one beam, a second beam with small angular diversity (i.e., roughly parallel to the first beam) may be weighted lower because it provides relatively little additional information about the location of the touch event beyond what the first beam provides. Conversely, a second beam which has a high angular diversity relative to the first beam may be given a higher weight in determining where along the first beam the touch point occurs.
Another factor for weighting beams is position difference between the emitters and/or detectors of the beams (i.e., spatial diversity). Usually, greater spatial diversity is given a higher weight since it represents “more” information compared to what is already available.
Another possible factor for weighting beams is the density of beams. If there are many beams traversing a region of the active area, then each beam is just one of many and any individual beam is less important and may be weighted less. Conversely, if there are few beams traversing a region of the active area, then each of those beams is more significant in the information that it carries and may be weighted more.
In another aspect, the nominal beam transmittance (i.e., the transmittance in the absence of a touch event) could be used to weight beams. Beams with higher nominal transmittance can be considered to be more “trustworthy” than those which have lower norminal transmittance since those are more vulnerable to noise. A signal-to-noise ratio, if available, can be used in a similar fashion to weight beams. Beams with higher signal-to-noise ratio may be considered to be more “trustworthy” and given higher weight.
The weightings, however determined, can be used in the calculation of a figure of merit (confidence) of a given template associated with a possible touch location. Beam transmittance/signal-to-noise ratio can also be used in the interpolation process, being gathered into a single measurement of confidence associated with the interpolated line derived from a given touch shadow in a line image. Those interpolated lines which are derived from a shadow composed of “trustworthy” beams can be given greater weight in the determination of the final touch point location than those which are derived from dubious beam data.
These weightings can be used in a number of different ways. In one approach, whether a candidate touch point is an actual touch event is determined based on combining the transmission coefficients for the beams (or a subset of the beams) that would be disturbed by the candidate touch point. The transmission coefficients can be combined in different ways: summing, averaging, taking median/percentile values or taking the root mean square, for example. The weightings can be included as part of this process: taking a weighted average rather than an unweighted average, for example. Combining multiple beams that overlap with a common contact area can result in a higher signal to noise ratio and/or a greater confidence decision. The combining can also be performed incrementally or iteratively, increasing the number of beams combined as necessary to achieve higher SNR, higher confidence decision and/or to otherwise reduce ambiguities in the determination of touch events.
In addition to intentional touches (also referred to as wanted touches) disturbing beams, unwanted touches may also disturb beams. Unintentional or unwanted touches are touches that a user does not want to be recognized as a touch. Unwanted touches may also be inadvertent, inadequate, aberrant, or indeterminate. For example, while interacting with writing or drawing application, a user may rest the side of their hand on the surface while writing with a fingertip or stylus. Consequently, the touch system may detect the palm touch and treat it as a touch event. Furthermore, if the user is resting their hand on the surface, the dorsal side of their fingers (e.g., the small and ring fingers) may also interrupt beams and cause additional touch events. In these cases, the palm touch and the dorsal touches are unwanted touches because they are not intended by the user to cause a response from the writing system. Once touches are classified by the touch system as wanted or unwanted, the touches may be reported to other systems such as an operating system or a PC controlling a display. In some embodiments, unwanted touches are not reported.
The classification of a touch as a wanted or unwanted touch may change over time. A touch may change from being an unwanted touch to a wanted touch (or vice versa) during a touch event. For example, a person may initially present a finger at an orientation which is not consistent with an intentional action and then roll their finger so that it shows the attributes of an intentional touch.
The front, back, or side of a hand will usually generate a touch which is large and elongated. Thus,
To classify touches as wanted or unwanted, a map of the touch events may be generated. The map may be referred to as an activity map and is a representation of touch activity detected by the touch system (e.g.,
A. Templates
The activity map may be formed using a set of a priori templates. In these embodiments, each template represents a region on the touch surface (e.g., see
A template value may be calculated by aggregating the transmission values Tjk of beams associated with that region. A template may be determined to be active if a proportion of the aggregated beam values Tjk have changed by more than a threshold amount (e.g., relative to beam transmission values Tjk in the absence of a touch event). The proportion of aggregated beam values Tjk may be specified to include all beam values associated with that region or specified to include a smaller subset of beam values Tjk associated with that region. The proportion of aggregated beam values Tjk may include any beams with beam values Tjk that have changed by more than a threshold amount or the subset may include a specified subset of beams. For example, the proportion of beams includes beams emitted from each side of the periphery. In another example, the proportion of beams includes beams with high angular diversity (e.g., three beam paths run at 60-degree angles to each other). Alternatively, a template may be determined to be active if the mean or average of the beam transmission values Tjk have changed by more than a threshold amount.
The threshold for determining whether a beam value is disturbed (e.g., blocked) by a touch event is preferably set at a level which is above the noise levels of the beam transmission value Tjk (meaning a beam is considered disturbed if the beam transmittance received by the detector drops to a level that is unlikely to be caused by noise alone). However, if touch events do not strongly affect the beams, template processing may be performed by using thresholds which are within the noise level of the transmissions values Tjk. In some cases, beam transmissions Tjk exceed the threshold solely due to noise, but false positives can be avoided by specifying that a template is only deemed to be active when a certain proportion of beams in the template are above the threshold because the probability of the specified proportion of beams exceeding the threshold due to random fluctuations is low.
Templates that are deemed to be active as a result of noise or other spurious activity can also be eliminated using rules of temporal or spatial continuity. With regard to temporal continuity, noise-induced activity is typically transient. As such, a template giving an active result due to noise in one computation associated with beam data at a time T1 is unlikely also to give an active result in successive computations for beam data at times T2, T3, etc. Thus, rules, heuristics, or other types of constraints may be placed on templates such that templates are only considered active if they are active for a threshold number of beam data sets within a determined time window. An additional or alternative constraint can mandate a template to be active for a contiguous set of beam data sets and/or mandate that templates near one another be active for a contiguous set of beam data sets (e.g., to allow for fast motion of a touch over successive scans)
Spatial continuity rules may be applied to eliminate templates that are incorrectly deemed active due to noise or spurious activity. Since templates deemed active due to a touch event are typically near or adjacent to other active templates, spatial continuity may be applied by evaluating templates which are located close to one another. Generally, the templates within 2 mm of one anther are grouped together although, the threshold for determining if templates are close may depend on the smallest contact size to be detected and the size of the template regions. For example, templates within 4 mm of each other are grouped together if the touch system is intended to detect fingers and the size (e.g. circumference) of the template regions is 2 mm. For example, an individual active template may be declared inactive if no nearby templates are also active. In some embodiments, morphological image processing methods are applied. For example, template results are treated as pixels in an image and morphological dilation and erosion are performed to affect a morphological closing function which removes small or isolated areas of activity.
In various embodiments, the area of each region is based on a minimum size of a touch which is to be identified and classified. For example, if the intention is to differentiate styli and fingertip touches from larger touches, then the template size can be large. This can be helpful in reducing the processing workload associated with classification. Conversely, applying templates to small regions of the sensing surface may use more computational resources, but can result in detailed representations of the touching objects in the activity map. In particular, a region size which is equal to the width the optical beams may allow for the smallest touches to be detected. For example, a touch system with optical beams with a width W of 4 mm can use a region size (i.e., would include beams with centerlines which are within a radius) of W/4, which is 1 mm in this example. To apply such a template, the beam transmissions Tjk for all beams with centerlines passing within a 1 mm radius of the template center can be processed to give a single result for that template, such as active/inactive. A complete set of templates covering the sensing surface at a distance of 1 mm center-to-center would ensure that a small (˜2 mm diameter) touch in any location would cause at least one template to respond.
Various settings for beam and template activity criteria may be used to generate multiple activity maps. For example, if multiple thresholds are established for determining active templates, an activity map can be formed based on touches which strongly affect beams (e.g., forceful touches) and another activity map can be formed based on touches which weakly effect beams (e.g., less forceful touches). This may be advantageous because intentional touches are generally more forceful than unwanted touches. For example, in optical waveguide touch systems, unwanted dorsal finger touches do not typically generate large changes to beam transmission coefficients Tjk compared to ventral finger touches due to lower applied force and lower sebum levels on the dorsal side skin surface. In another example, template sizes can be varied to generate multiple activity maps. For example, since unwanted (e.g., palm and forearm) touches tend to be larger than wanted (e.g., fingertip and stylus) touches, a first activity map with large template sizes may be generated to classify unwanted touches (smaller touches will not disturb enough beams to activate the larger templates). After the larger touches are identified, a second activity map with smaller template sizes may be generated to identify the smaller touches.
B. Template Clustering and Beam Analysis
Once templates are determined to be active or inactive, groups of active templates may represent touch events. Thus, active templates may be gathered into clusters to represent touch events. As a result, touches and touch characteristics can be recognized, determined, and reported. In some embodiments, morphological methods, such as shape matching are used to cluster active templates. Morphological analysis may also be applied to estimate touch characteristics such as size, orientation, degree of concavity, compactness, circularity, and shape factors (such as the aspect ratio) of the clusters.
To illustrate template clustering,
In another example,
C. Identifying Touch Types and Touch Groups
After an activity map is formed, touch types of touches on the activity map may be identified. A touch type of a touch event describes the object causing the touch event, a shape of the touch event, or a size of the touch event. Examples of touch types include finger-tip, finger-dorsal, finger-ventral, eraser-small, eraser-large, hand-side, stylus-type1, stylus-type2, object-triangle, object-square, object-equilateral triangle, arrow-pointing left, arrow-pointing right, arrow-at 45 degrees, forearm, sleeve, etc. As described below, touch types may be classified by a machine learned model. Touch types may also be classified based on the touch characteristics (e.g., identified in the activity map) of the touch events because touch objects typically have consistent touch characteristics. For example, fingertip touches are typically circular, dorsal touches are typically oval, and sleeve touches are typically triangular. In another example, although dorsal, palm, and forearm touches may have similar shapes, forearm touches are typically larger than palm touches and palm touches are typically larger than dorsal touches (e.g., type are assigned based on predetermined size ranges for each type).
Touch types may also be determined by analyzing beam transmission Tjk values. Touches by one or more touches types may disturb beams such that the touches are identifiable. For example, ventral finger touches on an optical waveguide sensor typically creates lower transmission values Tjk than dorsal touches due to higher sebum levels on the ventral skin surface. In some cases, styli and other instruments, such as erasers, disturb beams in a recognizable pattern. For example, a stylus is designed so that it disturbs beams from one direction differently than beams from an orthogonal direction. Thus, touches from styli may be classified as stylus touches by analyzing the transmission values Tjk of the disturbed beams.
In some embodiments, after an activity map is formed, touch events are categorized into groups. Touches may subsequently be classified as wanted or unwanted according to their assigned groups. Touches may be grouped according to touch types. For example, all palm touches are grouped together. Touches may be grouped according to their proximity to other touches. For example, all touches within a threshold distance from a first touch event are grouped together. Touches may also be grouped according to touch characteristics. For example, touches of similar size and/or shape are grouped together. Since intentional fingertip and stylus touches are typically circular and small, all circular touches with a diameter below a threshold may be grouped. In some embodiments, touches are grouped according to combinations of touch events. For example, groups are formed such that each group includes a fingertip and a palm touch. As indicated by the previous examples, groups may be formed according to combinations of criteria. For example, dorsal touches and palm touches near a stylus or fingertip touch are grouped together. In another example, dorsal touches near a palm touch are grouped together. These examples are further described with reference to
In some embodiments, contextual information is provided to the system. Context information may include a list of touch events a user may perform while interacting with the surface. In some embodiments, context information provides a list of touch events known by an application running on the touch system. Context information may also provide the type, size, and orientation of the touch system. For example, context information provides typical hand gestures that a user may perform when interacting with the touch surface. Thus, touch types and touch groups may be determined based on the typical hand gestures. For example, a palm touch may be shaped differently when a user is writing with a stylus compared to typing on a keyboard.
After an activity map is formed, touches can be classified as wanted or unwanted based on several methods. While the following methods are described in the context of an optical touch-sensitive system, these methods are not limited to being performed by an optical touch system (e.g., a capacitive touch system may use these methods to determine wanted and unwanted touch events). These methods can be performed individually or in combination. For example, a second method is performed to validate results from a first method. In another example, a first method is effective at classifying a first set of touch events and a second method is effective at classifying a second set of touch events.
While the following methods are explained in contexts where fingertip, stylus, dorsal, and palm touches are common, the methods are not limited to these touch types. Other touch types such as object touch types can also be classified. For example, these methods may be useful for identifying shapes in an educational application, such as recognizing a triangle as distinct from a square, or differentiating between an equilateral triangle and an isosceles triangle.
A. Contextual Information
The general classification of touches as wanted or unwanted may be dependent on received system context information and the configurations of touches or gestures present at a given time. An unwanted touch in one context may be a wanted in another context. For example, a palm touch on the surface is an unwanted touch if the user is writing on the surface with a stylus, but the same palm touch may be a wanted touch if the touch is intended to erase on-screen graphics. In some embodiments, context information indicates the system or application only accepts touches with specific touch types or characteristics. Thus, any touches other than the specified types and characteristics may be classified as unwanted. For example, a heads-up display (HUD) only accepts single touch events that are generally circular in shape. As a result, multiple touches and touches with noncircular shapes are classified as unwanted touch events. In some embodiments, context information includes user intent. For example, touch input may identify a desired user function (e.g., an erase function). Thus, unwanted touches can be determined based on the user intent. For example, if the system receives input indicating a user will perform typing gestures on a keypad, touches larger than a key on the keypad or between keys are classified as unwanted.
B. Machine Learning
Another approach to differentiating wanted and unwanted touches is to apply machine learning methods, such as support vector machines, random forest classifiers, or neural networks, to the activity map. Generally, there are two phases to such an approach:
The first phase is an a priori process of training a machine learning model using a population of data sets. Each data set A (e.g., an activity map, a set of templates, or a set of beam transmission values Tjk) is presented to the learning system along with an indication I as to whether the data set represents wanted touches, unwanted touches, neither, or both. Additionally or alternatively, the indication I includes touch types, touch characteristics, or touch groups that are present in the data set. The indication I may be provided by a human operator or an automated process. In an example automated process, the indication I is determined from additional sensors (e.g., an image capturing device) in a training touch system. In some embodiments, additional context input C is provided to the machine learning model during training.
In the second phase, the trained model is used to classify touches based on real-time data from a user interacting with the touch system. Depending on the training process, the model may classify individual touches or groups of touches as being wanted or unwanted. In some embodiments, the trained model classifies touch types (e.g., finger-dorsal) and subsequent processing determines whether a touch is wanted or unwanted. If the model is lacking sufficient data, such as context information, the model may classify one or more touches as “unknown.”
C. Predefined Rules
Touches may also be classified as wanted or unwanted according to a predefined set of rules. Often these rules are based on touch types and touch characteristics and may further be based on contextual information. For example, in a writing application, only fingertip and stylus touches are classified as wanted touches. In these embodiments, all other touches may be grouped together and classified as unwanted touches. Additional or alternative rules for a writing application include a rule specifying that touches smaller than a threshold size are wanted touches and a rule specifying that in a given context sleeve and forearm touches are unwanted touches. In some embodiments, touches within a threshold distance of another touch are ignored. For example, if a touch object (other than a finger or a hand) is identified, finger touches near the touch object are classified as unwanted, presumably because a user's fingers are unintentionally disturbing beams while holding the touch object.
Supplementary information from beam analysis and template clustering may also be used to classify touches. Templates representing an entire touch event (instead of a cluster of templates) can also be applied to the beam data. The shape and size of a cluster of active templates representing a touch can be used as a first selection criterion for the potential touch templates to represent an entire touch event. The quality of the fit to the potential touch templates based on the beam transmission values Tjk can be used to determine the classification of the touch event. Thus, a touch which does not fit any template may be classified as an unintentional touch. For example, if a cluster of active templates form an oval shape, the beam values Tjk may be compared dorsal touch templates. If the set of beam values Tjk for the cluster match a dorsal template, the touch may be classified as a dorsal touch. Conversely, if no match is found (e.g., within a confidence or matching threshold) the touch may be classified as unwanted touch.
One or more rules may be combined to form a decision flowchart. An example flowchart is illustrated in
In the embodiment shown, the method begins with a touch system detecting 2005 one or more touch events. As previously described, in response to the touch events, an activity map is generated 2007. Using the activity map, the touch system determines whether the touch is a dorsal touch 2010. A dorsal touch is touch event caused by a dorsal side of a finger. A touch may be determined to be a dorsal touch based on its size and shape. Typically, dorsal touches are oval and smaller than palm touches yet larger than fingertip or stylus touches (e.g., touch 1720).
If the touch is determined to be a dorsal touch, the system waits 2015 a predetermined period of time. Afterwards, the system determines 2020 whether the touch size and shape are constant. If the touch event continues to change in size or shape, the system waits 2015 another predetermined period of time until the size and shape are constant. Waiting may confirm that the touch not a result of noise. Waiting may also confirm the touch is not a larger touch that has not made full contact with the surface yet (e.g., the touch initially appears to be a smaller touch until the object fully contacts the surface). Furthermore, if the size or shape of the touch changes larger than a predetermined threshold, the touch type of the touch may be re-determined.
If the size and shape of the touch is constant, the system determines 2025 whether a palm touch is nearby (e.g., within a threshold distance on the touch surface).
If a palm touch is nearby, the dorsal touch and the palm touch are grouped 2030 together. As a group, the touches may be considered a single touch. In some embodiments, if other dorsal touches are detected nearby (e.g., a threshold distance away from the first dorsal touch), the other dorsal touches are included in the group (e.g., group 1750 is formed). Depending on the context, since the group includes one or more dorsal touches and a palm touch, the group is considered an unwanted touch (unless the system is configured to respond to such an arrangement of touches).
The system determines 2035 if the dorsal touch is located to the right of the palm touch. If the dorsal touch is located to the right of the palm touch, the system determines 2040 that the group of touches is from a left hand. As described with reference to
Referring to
If the touch is not a fingertip touch, it may be determined 2135 whether the touch is another touch type. If the touch is small, it may be a stylus or other instrument. Alternatively, the touch can be a contact which is above a maximum size threshold and, thus, is classified as being excessively large. If the touch is not identified as belonging to any known touch types, the touch can be classified as having an unknown type, and most probably deemed to be unwanted. As mentioned above, the touch may later be re-categorized as a known touch type once it has landed and stabilized (for these reasons, the classification of touches may be delayed for example on the order of tens of milliseconds).
Similar to steps 2015 and 2020, the device 100 waits 2010 a predetermined time period before determining 2115 whether the touch size and shape are constant. If the size and shape of the touch are constant, the system determines 2120 whether a palm touch (or a dorsal touch) is nearby (e.g., within a threshold distance on the touch surface).
If a palm touch is not nearby, the fingertip touch is classified as a wanted touch and the touch is reported 2125. In cases where intentional fingertip touches are common, the system may be designed such that fingertip touches pass through the method to step 2125 quickly compared to other method steps.
If a palm touch is nearby, the fingertip touch and the palm touch are grouped 2130 together. Depending on the context of the touches, the group may be a classified as wanted or an unwanted touch. Thus, if context information is not already known, context information may be retrieved to determine whether the group is a wanted or unwanted touch.
Referring to
If no additional dorsal touches are found, the dorsal touch is classified as a wanted touch and the touch is reported 2210. For example, the user is placing the dorsal side of their finger to perform a swiping gesture (e.g., to move an image or perform page turn function).
If additional dorsal touches are nearby, it is determined 2215 whether the additional dorsal touches are similar to the dorsal touch. Similarity may be based on touch characteristics. For example, if the additional dorsal touches have similar orientations and sizes (e.g., based on the long axes of the touch shapes), then the additional dorsal touches are classified as similar to the dorsal touch.
If the additional dorsal touches and the dorsal touch are similar, the touches are grouped 2220 together and reported. Depending on the system context, the group of touches may indicate that a hand is parallel to the touch surface and the dorsal side of several fingers are intentionally on touch the surface. For example, the hand is forming a gesture for moving an on-screen image.
If the additional dorsal touches and the first dorsal touch are not similar, the touches are also grouped together 2225. This can indicate that the dorsal side of several fingers are touching the surface but the hand is not parallel to the surface (e.g., touch 1530). For example, the hand is in a writing position on the touch surface. Thus, the group is classified as an unwanted touch.
After the dorsal touches are grouped together, it is determined 2230 whether the size of the dorsal touches increases from left to right. This may be based on the area of each touch or another shape parameter such as the length of the long axis of each touch. If the size of the dorsal touches increased from left to right, it is determined 2240 the group is from a right hand. If the size of the dorsal touches increases from right to left, it is determined 2235 the group is from a left hand. Similar to steps 2040 and 2145, a wanted touch on a writing side of the group may be expected and other touches within a threshold distance from the group can be classified as unwanted. In some embodiments, this process is generalized. If the dorsal touches are arranged such that sizes of the touches increase in size along a direction along the touch surface, a wanted touch is expected near the smallest dorsal touch along the direction, and other touches near the group may be categorized as unwanted touches and subsequently ignored.
The processing methods described above may be applied in whole or in part through distributed processing processes, such as remotely located processing nodes. This may be advantageous since the complexity of touch formations supported by these methods are likely to be less commonly encountered than the typical groups of fingertip or stylus touches. For example, a touch sensing system in an educational environment might be tasked with recognizing geometric shapes presented to the sensing surface and indicating the detected shape to a student. This shape recognition can be done, for example, using trained machine learning systems in the cloud. Such a scheme can be extended to high levels of complexity, allowing formations of shapes at various orientations and scales to be recognized or interpreted by resources which need not be contained in the hardware directly associated with the system. In this way, the cost of the hardware can be reduced. By exposure to training data from a population of multiple touch systems, a remote learning system can be trained and improve its performance. Also, central administration of the touch analysis processing facilitates the addition of new touch types and formations to the remote system. New touch types can include geometric shapes, or special objects or instruments with distinctive optical properties such as those disclosed in U.S. Pat. No. 9,719,976.
Using a combination of local and remote processing systems may mean that latency delays may be small in response to frequently encountered (and locally processed) touches such as fingertips, while uncommon touch types or formations may be processed remotely. Remote processing is likely to incur additional latency, but that may be tolerable when the touch types or formations are uncommon. For example, the increased latency may be offset by an increase in processing power that is available remotely, reducing the overall processing time.
The determined touch characteristics (e.g., size and shape) of a touch may change over time. For example, the size of a touch event from a stylus decreases in size due the user applying less force to the stylus. Additionally, new touches may be detected near a given touch. As a result, the touch types of these touch events may be reclassified e.g., as long as the touches are detected. A touch type revision scheme can allow for the touch type to be changed. For example, if the type attributed to a touch is changed, the activity caused by the touch in its previous type can be revoked (e.g., undone), and the touch trajectory can be retraced with the new touch type.
In a case where a set of previously separate touches are revised to belong to a single touching object, the activity caused by the reported trajectories for the touches can be revoked and combined into a single trajectory. Then, the single touch trajectory (or a new recalculated trajectory which represents the motion of the touching object) can be retraced as the new touch type. Alternatively, the activity caused by the reported trajectories for the touches can be revoked, and all of those touches released. Then, a new touch can be generated with the new touch type, and the trajectory of one of the released touches (or a new recalculated trajectory which represents the motion of the touching object) can be traced by the new touch.
In some cases, separate touches can interfere with one another, even if the touches are similar in size and optical absorption. For example, beams passing through a stylus touch can also pass through a larger nearby palm touch. As such the beam transmission values Tjk for the beams disturbed by the stylus touch may also be affected by the palm touch. This may disrupt the location estimate for the stylus touch.
In these cases, one or more interfering touches may be temporarily classified as unwanted touch events until a location (or other touch characteristics) of a touch event is determined. For example, if a palm touch is interfering with estimating a location of a stylus touch, the palm touch is considered an unwanted touch until the location of the stylus touch is determined (even if the palm touch is later determined to be a wanted touch).
An activity map may enable separate touches to be identified and the underlying beam data used to reduce interference interactions between touches. First, the touches as represented by clusters of active templates can be separated using image processing methods such as the recursive grassfire algorithm. Then, decomposing the templates for each touch into the underlying beams associated can provide a beam list for each touch. Finally, the beam list data can be used to reduce the disturbance between touches:
In some embodiments, this problem is solved by identifying beams for a touch that are not shared with any other touches. Thus, the identified beams may be used to determine the location and other touch characteristics of the touch. Using only the beams which are unique to a touch can eliminate or reduce the effects of other touches.
In other embodiments, the system may identify beams impacted by a touch event that are shared with other touches. For each shared beam, the system estimates the contribution of each touch to a change in transmission values Tjk caused by each of the touches along that beam. This estimation can be derived by applying a loss per unit distance rule and tracing the path length for each beam through each touch (for example by counting how many template regions the beam passes through in each touch). The loss per unit distance can be estimated by taking a percentile of the change in transmittance values Tjk for a population of the beams passing through a touch, or by identifying unshared beams (e.g., beams which are only affected by the touch in question) and dividing the transmission Tjk loss of those beams by their path length through the touch. Another way to estimate the contribution of each touch on the change in transmission on a shared beam is to identify unshared beams which should be affected by each touch in a same or similar manner as the shared beam. For example, a shared beam passing through the center of a circular touch should experience similar transmission Tjk loss from that touch as an unshared beam which also passes through the center of the touch (e.g., from a different direction).
When the effect of each touch on a shared beam has been apportioned to the touches, new beam values Tjk(1), Tjk(2) . . . Tjk(N) can be calculated for each beam, where the values are the transmission value for the beam segment passing through touches 1 to N. Thus, these calculated transmission values Tjk(x) for shared beams can be combined with the measured transmission values Tjk for unshared beams, and used to determine the location (and other touch characteristics) of a touch.
A. Introduction
In various embodiments where multiple touches can be detected simultaneously, various advantages may be realized by compensating for the effects of one touch event on the detection of other touch events. For example, FTIR (frustrated total internal reflection—where the sensing light propagates in a waveguide material and is disturbed or frustrated by touches (e.g., see
In various embodiments, optical emitters are enumerated from 1 to Ne, and photodetectors are enumerated from 1 to Nd. A beam is defined by a pair including an optical emitter ej and a photodetector dk, where j and k are indices of the optical emitter and photodetector, respectively. Beams are enumerated form 1 to Nb, where Nb is generally not greater than Ne*Nd. A mapping from emitter-detector index pair (j, k) to the corresponding beam index n is established.
Beam power for a given beam is defined as the optical power reaching its photodetector. Beam transmission coefficient Tjk is an indication of the difference between the instantaneous beam power and a reference beam power (e.g., a ratio of the instantaneous and reference beam powers). The reference beam power is the power measured before any touch is applied; the reference beam power may be referred to as beam power baseline, or simply beam baseline. Beam transmission loss may be defined as (1−beam transmission coefficient Tjk) and is associated with touch absorption. A more absorbent touch generally gives rise to a larger beam transmission loss for beams propagating below the touch. Touch absorption is related to the size of the object in contact with the sensor, how efficiently light passes through it, its refractive index, its reflectance and (for FTIR sensors) the quality of contact's optical bonding. “Beam response” and “beam transmission loss” may be used interchangeable herein. The beam response of a beam measured in the absence of touch events may be referred to as the baseline beam response.
A touch response pertains to a touch event and a set of beams. A touch response indicates beam responses for the beams in a given set of beams when the touch is present. The set of beams may include the entire system's beam population, but more commonly it refers to beams that are in the vicinity of the touch, i.e. the beams attenuated by the touch. Herein, a ghost refers to a set of beam responses that may resemble a touch response, but do not come from an actual touch. In one example, the system is able to recognize and ignore ghosts. Additionally, a beam can be categorized as a shared beam if it is affected (i.e. non-negligibly affected) by at least two touches.
One difficulty in using beam response to a touch is that the response may be influenced by other factors, such as other interacting touches and contamination residual from prior touch activity. Beams influenced by multiple contributors can degrade system performance in multiple ways: i) a touch response may not be detected; e.g. the touch response is small when compared to other neighboring touch responses, ii) an estimated touch position may be inaccurate; e.g. beam transmission loss caused by neighboring touches confuse the position estimation, iii) a true touch may be classified as a ghost and not reported, or a ghost may be classified as a true touch (the beam responses to neighboring touches confuse the ghost classifier). The third classification error may also affect touch type classification, where the objective is to determine which kind of object is touching the sensor (finger, stylus, eraser, etc.).
Another class of disturbance, called self-disturbance, arises where previous activities associated with a touch negatively affect its present touch tracking. One example is a finger leaving a contamination trail (e.g. finger sebum or food residue) behind. Contamination trails may contribute to beam response in ways similar to touches. Hence it may be useful to limit tracking degradation of a given touch caused by other touches (present and previous activities) and by that same touch (previous activities).
B. Overview
Each beam response can result from multiple contributors, such as other touches affecting that beam, or virtual touches from contaminants at past touch locations. In various embodiments, a linear optical touch system, such as one based on FTIR, each source (e.g., a touch or virtual touch) contributes additively to the beam response of a beam. The global beam response is the sum of beam responses of each source taken in isolation. In the linear case, active compensation for a given source amounts to subtracting predicted beam response contributions associated with other known sources. This active touch compensation strategy is referred to herein as “linear touch compensation.”
Another approach may be more convenient for non-linear systems, such as ones where beam responses can saturate in the presence of a touch. This is typical with OTS systems, where touches physically intercept beams and can give rise to touch responses close to 1 (i.e., where the detected beam intensity is reduced to zero or almost zero). Owing to this non-linearity, the previously described separation of the beam response into separate touch contributions (linear touch compensation) may not be useful. In these non-linear cases, it may be advantageous to use a different approach that tracks beams affected by each touch, and when computing attributes of any touch, ignores the beams affected by other touches. This active touch compensation strategy is referred to herein as “full interference rejection.” It is worth noting that full interference rejection may be used not only for non-linear systems s, but also with linear systems, such as an FTIR system. Similarly, in some embodiments, linear touch compensation may be used for non-linear systems.
In some embodiments, when a contamination deposition is tracked for the purpose of compensation, a second list named virtual touch list is maintained. This list holds a list of virtual touches. The number of virtual touches used to model the finger oil or other contamination trails (as will be described later) is denoted by Nv. The virtual touch list can be separated from the tracked touch list, or the lists can be combined, with virtual touches marked as such.
As shown in
C. Touch Model Calculator
In some embodiments, a dedicated process called a touch model calculator 2310 calculates the beam response contributions from each touch to each beam. This calculation can take place every frame (the time taken to complete all optical emitter activations and associated detector acquisitions), for a fraction of a frame (subframe), or according to any suitable time parameters. In the case of a subframe, the contributions may be calculated for the set of beams activated during that subframe. For example, the emitter activation cycle can be divided into four parts of similar duration, in which case the subframe is called a quarter-scan. Described herein, a frame can refer either to a full frame or a subframe.
In a first phase common to both active compensation approaches (linear touch compensation and full interference rejection), the touch object is modelled (for example by a given shape), and the parameters associated with the model are estimated. In some embodiments, the touch model is that of a disc shape, and its parameters are radius, position, and touch strength. Other models with their appropriate parameters are possible, and those include, but are not limited to, shapes such as ellipses, rectangles (e.g. erasers), etc. Touches can also have more complicated shapes. For example, when a palm or a hand is placed on the screen, this may be modeled as a superposition of touches of simpler shapes mentioned above or approximated by a pixel-based representation of the touch outline.
In some embodiments, the model parameters are estimated using a set of beam responses from a current frame. In alternative embodiments, the parameters have already been calculated as part of the tracking loop and are recirculated from the previous frame. The second approach is illustrated in
C1. Template Estimation
A template representation of each touch (e.g., as described above in Section V. Activity Maps) typically understates the extent of the touch boundary since some of the inactive beams near the touch will have been slightly affected by it, but not sufficiently to be deemed active. To compensate for this effect, the active template region of the touch can be enlarged using a dilation method.
In some embodiments of template representations of touches, the touch model calculator 2310 applies a reverse activity mapping process which determines the beam responses for each touch based on the template representation and each beam incident upon it.
In some embodiments, an activity map is determined from the template representations. Shared beams may be identified using the activity map. After the shared beams are identified, the beam responses are compensated according to the linear touch method or the full interference method. A new activity map may be determined based on the compensated beam responses. Among other advantages, the new activity may more accurately describe the touch events on the surface.
C2. Linear Touch Compensation—Phase 2
In a second phase of the linear touch compensation approach, beam responses are calculated for each touch using the current touch model calculated in phase 1. In various embodiments, the set of beams interacting with the touch is defined, referred to as the touch beam list. For example, where a disc touch model is used, this list is based on the touch size and the distance from the touch center to each beam. The touch model is used to calculate the normalized model beam response for each beam in the touch beam list. The normalized model beam response corresponds to the touch of a given size and position with a unit strength, and is based on an analysis of the optical setup used in the touchscreen, including beam positions, beam widths, etc. The normalized model beam responses may be scaled by touch strength, where touch strength is a factor which aligns the amplitude of normalized model beam responses to the measured beam responses.
In an FTIR sensor, touch strength can relate to the degree of optical bonding between the touch and the waveguide. In an OTS sensor, touch strength can relate to the optical absorbance of the touch. One method to obtain the value of touch strength is to minimize the squared difference of scaled normalized model beam responses with measured beam response for a defined set of beams. In some embodiments, this set of beams, referred to herein as unshared touch beams, includes beams that are affected only by the considered touch and no other touches.
In some embodiments of the linear touch compensation approach, the touch model calculator 2310 computes two lists, Ttouchmodel and Tmodel. The list Ttouchmodel has, for each tracked touch i, a full beam list Ttouchmodel(i), whose entry Ttouchmodel(i,n) at index n contains the beam response of beam n predicted by the model parameters for touch i. For example, the beam responses are the normalized model beam responses described above. Beams not affected have a contribution of zero. The other list, denoted Tmodel, is common to all tracked touches and contains the list of all beams, with touch response contributions summed on a per-beam basis. In other words, Tmodel contains Nb elements, where element Tmodel(n) corresponds to the beam with index n and contains the sum of Ttouchmodel(i,n) over i:
Compensation of tracked touches may be performed by element-wise subtracting the model beam responses contained in the Tmodel list from the measured beam responses contained in the list Tmeasured:
The compensated beam response list contains items that may not be modeled by the tracked touches. Thus, in
C3. Full Interference—Phase 2
The full interference rejection approach has a somewhat simpler second phase. For each touch in the tracked touch list, the corresponding model parameters are used to mark the beams that the touch attenuates or interacts with. Beams can be marked as attenuated based on their distance from the touch, their width, and the size and shape of the touch.
In some embodiments of full interference rejection, the touch model calculator 2310 computes two lists, Ttouchmodel and Tmodel. The Ttouchmodel list contains, for a tracked touch i, a beam list, where an entry Ttouchmodel(i,n) at index n contains 1 if touch i interacts with beam n, and 0 otherwise. The Tmodel list is an element-wise sum of the Ttouchmodel lists over all values of the touch index i.
Note that for any beam n, Tmodel(n) contains the number of touches affecting beam n; shared beams are beams with Tmodel(n)>1. Compensation of tracked touches with the full interference rejection approach may include ignoring (or forcing to zero) all measured beam responses for beams affected by any of the tracked touches. The computation may be done as follows:
D. New Candidate Detection
In some embodiments of a touch tracking loop, known touches are tracked and maintained (e.g., recirculated in the tracking loop) in the touch tracking list unless a touch departure is detected or a touch gets marked as a ghost. The new candidate detector 2320 provides a list of new candidates that includes touches which have just landed. New candidates are combined with existing touches, if any, confirmed during the previous frame and maintained in the tracked touch list. The new candidate detector 2320 can use measured beam responses Tmeasured, and detect new touch candidates based on various methods, such as increased responses for beams passing through a particular region of the touch surface.
In some embodiments, the new candidate detector 2320 uses a set of compensated beam responses Tcompensated instead of the measured beam responses Tmeasured. In so doing, the new candidate detector 2320 operates on data related to changes and new touch events rather than data which is already captured in the tracking touch list. This can reduce the computational workload associated with new touch detection.
D1. Instantaneous Beam Response
In some embodiments, instead of feeding to the new candidate detector 2320 the measured beam responses Tmeasured or the compensated beam responses Tcompensated, the new candidate detector 2320 receives beam responses calculated with a different baseline beam response. As described previously, the baseline beam response is the beam response measured in the absence of any touch events. However, due to the complexity of modeling and tracking contamination through the virtual touch list, beam responses may be measured by using the beam response for M frames (for example, M=2) in the past as the baseline beam response value. This baseline is called an accelerated baseline and the beam response calculated with the accelerated baseline is called instantaneous beam transmission loss or instantaneous beam response. The instantaneous beam responses computed with the accelerated baseline may be stored in Tinstantaneous. When a touch moves, depositing a contaminant behind, the apparent activity due to the remaining contaminant disappears quickly in the Tinstantaneous array, and this apparently implicit touch compensation happens at a low cost. In some embodiments, a single accelerated beam baseline is used. For example, the accelerated beam baseline is the average beam response of the beams for M frames in the past. In other embodiments, multiple accelerated beam baselines are used. For example, each beam is associated with a different accelerated beam baseline.
Location, size, strength, and confidence are typical attributes of a touch. The touch system 100 may describe touch attributes with desired fidelity to the host system, recycles validated touches back into the tracking loop, and discards touches that took off (for example based on a combination of strength and confidence) or that are identified as ghosts (for example based on confidence, a ghost being a low-confidence touch).
Accuracy and linearity are measures of precision of the touch position tracking. For a given touch, the presence of neighboring touches can be seen as an interference that drives the estimated position away from its true value. Likewise, lack of accuracy in confidence estimation may result in ghosts being not properly identified and suppressed, producing lasting and visible spurious touches; similarly, touches can appear as ghosts even though they may be present on the screen.
E. Touch Attribute Estimator
In various embodiments of touch tracking system, a process called the touch attribute estimator 2340 is responsible for estimating touch parameters such as position, size, confidence, and strength. In some embodiments, the touch attribute estimator 2340 executes in three phases: touch beam selection, beam response compensation, and beam attribute computation.
E1. Touch Beam Selection
In the first phase (as previously described above), the touch attribute estimator 2340 selects a list of beams that are used to compute touch attributes for a given touch i. In some embodiments described below, the beam response values for the selected beams are adjusted to compensate the contributions of touches other than the one being analyzed.
The beams selected by the touch attribute estimator 2340 can be selected based on the touch size (computed in previous frame or taken as a default value for new touches) and the distance from the touch center (estimated in previous frame or in the new candidate detector 2320). The selected beams are not only those beams that may be affected by the touch, but also beams further away, which allows localizing the true current position of the touch, as well as its current size.
E2. Beam Response Compensation
In the second phase (as previously described above), the beam responses used for touch attribute estimation are compensated. There can be different compensation strategies. For example, one may be relevant for linear touch compensation while another may be used for full interference rejection. Regardless of which strategy is used, the desired result is to remove beam response contributions from touches other than the touch being analyzed (denoted as touch i) and leave the contributions of the analyzed touch unchanged. Among other advantages, fidelity of attribute estimation can be improved with the removal of contributions from other touches. This generally results in more precise touch positions and more robust ghost and other touch type classification.
E2a. Beam Response Compensation—Linear
In the section on computing touch contribution, we have seen that the Tcompensated list may contain beam responses in which the model beam response contribution from the touches have been removed. In order to perform touch attribute computation for a given touch i, the system can work with beam responses where the contribution of other touches has been removed (or at least substantially removed). Tattibute(i) denotes the beam list used for computing the attributes of touch i, in which the contribution of other touches has been compensated (e.g., removed). The beam list Tattribute(i) is obtained by adding the list Ttouchmodel(i) to the list Tcompensated:
E2b. Beam Response Compensation—Full Interference
With the full interference rejection strategy, a more extreme compensation of tracked touches' beam response contribution may be performed: beam responses of shared beams are removed. For touch attribute computation of a given touch i, the beams marked as affected by that touch and no other touches are added back. In some embodiments, the beam response values used for touch attribute computation are obtained from the measured beam responses by considering the lists Ttouchmodel(i) and Tmodel(i): the beams that are affected by no touches or only touch i are kept, and those affected by other touches are rejected. Thus:
E3. Beam Attribute Computation
In the third phase, compensated beam responses are used to compute the touch attributes that include location, size, confidence, and strength (e.g., as previously described above in Sections III. or V.). In some embodiments, touch event location is determined by using a goodness-of-fit algorithm to compare the beam responses with touch event template and selecting a template with the best fit or a fit within a threshold error. Confidence may be determined based on the goodness-of-fit for the selected template (e.g., the error of the goodness-of-fit). Strength may be determined by the scaling applied to the selected template to match the observed beam responses.
F. Virtual Touch Model Calculator
For compensation of self-touch contribution, a virtual touch model calculator (not illustrated) may predict beam responses from oil deposited at previous finger positions. When finger oil residue (e.g., sebum) is left over on a finger trajectory, the oily trace can be modelled by a single (virtual) touch having the shape of the oily trace. Note that a virtual touch may be considered a ghost touch. In some embodiments, the oily trace is modelled by a discrete set of simple circular touches. In the simplest case, the oily trace is modelled by a single circular virtual touch of appropriate radius and located at a finger past position.
To define the (virtual) touch model, the touch past attributes are stored in a list called tracked touch history. With the help of the tracked touch history, a list of virtual touches in created, called virtual touch list. This list is similar to the tracked touch list, except that it includes virtual touches such as those resulting from oil residue at past positions. Generally, these touches are not reported to the host system (although they may be in some embodiments).
In some embodiments, assuming the oily trace is modelled with Nv virtual touches, the virtual touch list is obtained by selecting Nv past touches (for example Nv=2 virtual touches in
In some embodiments, the seed position used to produce the virtual touch list is estimated independently of the virtual touch model calculator and touch attribute estimator 2340, given that the estimated touch position based on touch attribute estimator 2340 is negatively affected by contamination. In various embodiments, the seed position is obtained using a rough, possibly noisy, position estimator (not illustrated) obtained with a more robust method. In one such embodiment, the seed position is based on a position estimator (not illustrated) using the accelerated beam baseline. This method for obtaining the current finger position can track finger motion even when there is contamination, which may not be the case when using the regular transmission coefficient definition.
In a way similar to the touch model calculator 2310, the virtual touch models can be used to produce a set of activated beam list and a matching set of normalized beam responses. The normalized beam responses may be further scaled by virtual touch scaling factor so that scaled normalized transmission loss will match the beam response measurement on a subset of beams interacting with the virtual touch.
The end result is Tvirtualtouch(i,v,n), where n specifies the beam, i specifies the index of a tracked touch, and v specifies the virtual touch index (from 1 to Nv). For example, a virtual touch with index 1 may be the touch selected in the tracked touch history to be the most closely located at distance virtual distance from the current touch position.
G. Compensation Using Virtual Touch Model
The new candidate detector 2320 and touch attribute estimator 2340 can use virtual touch compensation. In order to isolate the beam responses of the touch under consideration from other virtual touches produced by that same touch, compensated beam responses can be applied to the processing blocks. More specifically, to detect new touches in the new candidate detector 2320, a list of fully compensated beam responses, Tfullycompensated, may be provided to the new candidate detector 2320, where both tracked and virtual touches are compensated. The compensation takes the following form:
Likewise, to estimate parameters of touch i in the touch attribute estimator 2240, the compensated beam responses Tfullattribute(i) may be fed to the new candidate detector 2320, with the touch model beam response being added back:
Thus, some embodiments relate to a method for classifying touch events on or near a surface, the surface having emitters and detectors arranged around its periphery, the emitters producing optical beams received by the detectors, where touch events disturb optical beams. The method includes the following steps: estimating a position of a current touch event; selecting one or more past touches, the one or more past touches selected from a tracked touch list based on a distance between the one or more past touches and the estimated position of the current touch event; determining whether the current touch event is a contamination touch event based on the one or more past touches; and responsive to determining the current touch event is a contamination touch event, ignoring the current touch event.
H. Touch Tracker—Merging Touches
Merging new candidates with established touches can be done in the touch tracker 2350, based on the annotated touch list consisting of both known touches (from the previous frame) and new candidates (from new measurements). This may involve identifying duplicate touches and selecting one of the two duplicate touches for recirculation in the tracking loop, as explained below. Touches presented by the new candidate detector 2320 are generally free of any contribution from other touches and virtual touches known in the tracking loop. In most cases, new candidates result from the landing of additional touches.
In some instances, however, a new candidate may represent a touch that is already present in the touch tracking list. This may happen, for example, when there is a discrepancy between the touch model and the effective touch response. In other examples, model discrepancies occur when the predicted position is different from the effective touch position—for example after a very large unpredicted acceleration. Position mismatches result in discrepancies between model with measurement. In this case, an additional new candidate may be detected, though it does not correspond to a new touch. In other instances, when contamination is heavy, smudging around the effective touch may be interpreted as a near-stationary touch. As a result, heavy contamination can give rise to a stuck touch, the stuck touch actually being the (stationary) contamination. This situation produces an additional new candidate (the moving touch) which is actually a touch already known to the system. The stuck touch is contamination, a non-legitimate touch which may be discarded, and the new candidate can then be associated to the known touch.
In various embodiments, in a first step, touch matching is performed. Distinction between the landing of a new touch and matching between a tracked touch and a new candidate can be based on various criteria, among others but not limited to, distance. In some embodiments, a new candidate and an existing touch are declared matched when their distance is below a threshold. For example, a threshold of 5 mm can be used.
In a second step, matched pairs of tracked touches and new touch candidates are fused into single touches. The fusion criteria may be based on a set of rules that use available attributes, such as prediction error, strength, and confidence, and with some established priority. For tracking, two positions may be merged. In some embodiments, the fused position is a weighted sum of the position of the tracked touch and the position of the new candidate, where the weights are obtained from a multivariate mapping from the two touches' strength, confidence, and distance values from the predicted position (i.e. prediction error). Generally, the weights are non-decreasing functions of strength and confidence, and non-increasing functions of the prediction error.
In some embodiments, a tracked touch with large size and strength attributes is discarded, as these values may be typical of contamination. For contamination discrimination, the fusion strategy can check attributes values and compare them to representative values of either contamination or a finger.
Merging position discrepancies linked to position prediction errors can be based on similar rules, such as reduced strength observed for an accelerating touch. In these embodiments, the new candidate is selected as a legitimate touch when its trajectory is consistent with large acceleration and when a reduced strength is observed. Given multiple rules in the selection process, rules can be prioritized.
One or more touch events are identified 2705. Each touch event has a shape and a size.
Based on the shapes and sizes, a touch type is assigned 2710 to each of the one or more touch events. Touch types may include a fingertip touch type, a stylus touch type, and a palm touch type. Assigning touch types may be performed by a machine learned model trained from data sets. Each data set includes information indicated a multiple touch events and labels classifying touch events as fingertips touches, stylus touches, and palm touches.
At least one touch event is classified 2715 as an unwanted touch event based at least in part on the assigned touch type of the at least one touch event. Touch events assigned as palm touch types may be classified as unwanted touch events. Classifying at least one group as a group of unwanted touch events may be performed by a machine learned model trained from data sets. Each data set includes information indicating multiple touch events and labels classifying touch events as wanted touches and unwanted touches.
In some embodiments, touch types further include a stylus touch type, a dorsal touch type, a forearm touch type, and a sleeve touch type. In these embodiments, classifying at least one touch event as an unwanted touch event includes classifying touch events assigned as dorsal touch types and sleeve touch types as unwanted touch events.
The method may include receiving context information including a list of touch types that can be classified as wanted. In these embodiments, classifying at least one touch event as an unwanted touch event is further based at least in part on the context information.
The method may include grouping the one or more touch events into groups. The grouping is based on the assigned touch types. In these embodiments, classifying at least one touch event as an unwanted touch event is further based at least in part on the grouping.
One or more touch events are identified 2805. Each touch event has a location and a shape. Subsequent to identifying one or more touch events, one or more of the touch events may be classified as a fingertip touch event, a stylus touch event, a dorsal touch event, or a palm touch event based on the shapes of the touch events.
The one or more touch events are grouped 2810 into groups. The grouping is based on the locations and shapes of each of the one or more touch events. The grouping may also be based on sizes and orientations of each of the touch events. The grouping may be performed by a machine learned model trained from data sets. Each data set includes information indicated a plurality of touch events and labels classifying touch events as wanted touches or unwanted touches.
At least one group is classified 2815 as a group of unwanted touch events based at least in part on the grouping. The group may be classifying may be performed by a machine learned model trained from data sets. Each data set includes information indicating a plurality of touch events and labels classifying touch events as a fingertip touches, stylus touches, dorsal touches, and palm touches. In some embodiments, the group of unwanted touch events includes a dorsal touch event and a palm touch event. In some embodiments, the group of unwanted touch events includes two or more dorsal touches.
A set of touch event templates are a priori determined 2905 for a group of expected touch events. Each touch event template represents a region of the surface and is defined by a corresponding set of at least two beams that would be disturbed by an expected touch event at the region.
Information indicating which beams have been disturbed by actual touch events is received 2910.
A subset of active templates is determined 2915 from the set of touch event templates. Each active template is a touch event template wherein the corresponding set of beams is disturbed by the actual touch events.
In some embodiments, determining the subset of active templates comprises additional steps. For each touch event template, beam transmission values for each beam in the corresponding set of beams are obtained. For each touch event template, it is determined whether at least a specified proportion of the transmission values exceed a predetermined threshold. For each touch event template, if the specified proportion exceed the predetermined threshold, the touch event template is included in the subset of active templates. In these embodiments, determining if the specified proportion exceeds the predetermined threshold comprises determining whether the specified proportion exceeds the predetermined threshold for a threshold amount of time.
An activity map is formed 2920 based on the subset of active templates, the activity map representing the actual touch events on or near the surface. The resolution of the activity map may be determined by a size of the regions represented by the touch event templates.
In some embodiments, the activity map is formed by clustering active templates into clusters based on the regions of the surface corresponding to the active templates. A first active template may be included in a cluster with a second active template if a first region of the surface corresponding to the first active template is no more than a threshold distance from a second region of the surface corresponding to the second active template.
The touch-sensitive devices and methods described above can be used in various applications. Touch-sensitive displays are one class of application. This includes displays for tablets, laptops, desktops, gaming consoles, smart phones and other types of compute devices. It also includes displays for TVs, digital signage, public information, whiteboards, e-readers and other types of good resolution displays. However, they can also be used on smaller or lower resolution displays: simpler cell phones, user controls (photocopier controls, printer controls, control of appliances, etc.). These touch-sensitive devices can also be used in applications other than displays. The “surface” over which the touches are detected could be a passive element, such as a printed image or simply some hard surface. This application could be used as a user interface, similar to a trackball or mouse.
This application claims the benefit of and priority to, U.S. Provisional Application No. 62/826,567, filed on Mar. 29, 2019, which is incorporated herein by reference in its entirety for all purposes.
| Number | Date | Country | |
|---|---|---|---|
| 62826567 | Mar 2019 | US |
