Patent Application
20160154021
The present invention relates to a device for calculation, with high time resolution, of acceleration of an object in motion from a speed measurement with low time resolution with associated quality measure of speed measurement; comprising means for estimation of current speed with a parametric model describing the dynamics of the motion; means for calculating an acceleration from the parametric model; means for calculating a quality index for said calculated acceleration from calculated quality of said parametric model and said quality measure of said speed measurement.
The introduction of so-called Smartphones such as iPhone and Android-based phones such as for example HTC Desire has increased the availability of the information technology—the functionality of the mobile phone has been multiplied from being a device for voice calls, to a device with a versatile field of application. Other devices with overlapping functionality comprise for example palm-pilots, tablets such as iPad and Android-based tablets such as Samsung Galaxy Tab P1000, notebooks, PC laptops or other general portable computer products where the functionality for the user may be adapted by downloading of computer programs from electronic market places such as App Store or Android Market or computer-readable media such as CD, DVD, USB-memory, hard drive, etc. The invention applies to personal electronics such as exemplified above, which for simplicity are given the collective name mobile phone, and in particular when this personal electronics is used in a motor vehicle during travel.
Modern mobile phones often have built-in receivers for satellite navigation systems. Several satellite navigation systems are in use such as for example GPS (United States NAVSTAR Global Positioning System), GLONASS (Russian Global Navigation Satellite System), Galileo (Europe), COMPASS (China)—which are gathered under the collective name GNSS (Global Navigation Satellite System). An example of GNSS with support from local positioning with the aid of the mobile phone systems is assisted GPS (A-GPS). In particular GPS of the different GNSS-systems has had a major impact and GPS receivers are nowadays found in a majority of mobile phones, in a great majority they have support for A-GPS. Also GLONASS is common nowadays.
The GNSS systems deliver information on current speed, position, direction of travel (heading), time, with associated quality measures through standardized protocols such as NMEA 0183 or through vendor-specific protocols; Trimble Standard Interface Protocol and SiRF Binary Protocol are two examples. Data from GNSS receivers are used in a variety of applications, for example car navigation systems, maritime navigation systems, traffic flow measurements, drivers log systems and fleet management. Data from GNSS receivers are also used for determining position for location-based services and functionality such as marking of digital photographs, location-based search services for market offerings, timetables and route lists, news services.
The vast majority of mobile phones with built-in GNSS receivers deliver data with 1 second intervals, i.e. 1 Hertz update rate. This is enough for the applications exemplified above, and is a result of demands for energy efficiency and cost efficiency that exist on this class of products.
Despite the above-mentioned limitation in data update rate, there is a need for using mobile phones for other applications than the applications they are originally intended for. Such an example is for detection of rapid speed changes. Rapid speed changes occur for example when a car driver performs heavy braking for the purpose of stopping the vehicle. Precisely detection of heavy braking may be used as a risk parameter when calculating an insurance premium for a vehicle based on driving behavior. A driver with a large number of braking maneuvers, measured over driving time or driving distance, may indicate a higher risk factor than a driver with a lower number of heavy braking maneuvers.
A car insurance premium for private cars is traditionally based on the classification of the vehicle owner and the vehicle in terms of vehicle type, driving distance, age, gender, geographical residence and number of damage-free years. These are by necessity blunt instruments for determining an insurance premium. For an expert it is obvious that similar calculation rules apply to other types of motor vehicles such as buses and trucks.
Premium calculation based on actual driving behaviour is on the market, for example the insurance company If's SafeDrive. Through active monitoring of the vehicle by technical equipment a premium may be calculated not only from the above-mentioned list of criteria, but also from for example
The mentioned technical equipment may be fixedly installed, or consist of a modern mobile phone, since a modern mobile phone is not only equipped with GNSS receivers but also with sensors such as accelerometer and gyro. If's SafeDrive is available as an app (computer program) for iPhone.
The invention is related to detection of strong acceleration and retardation. These are normally detected by means of an accelerometer which thus may be fixedly mounted in the vehicle, alternatively an accelerometer in a mobile phone which in turn is fixedly mounted in the vehicle in a holding device intended for the purpose. Fixed mounting is necessary since an accelerometer cannot separate the true acceleration of the vehicle from the force of the Earth's gravity. In order to accurately measure the acceleration of the vehicle by means of a fixedly mounted accelerometer the angle between the sensitivity axis of the accelerometer and the direction of the Earth's acceleration must be known with an accuracy of a few degrees.
Today most new mobile phones are equipped with Micro-Electro-Mechanical System (MEMS) accelerometers which in theory may be used for detection of strong acceleration and retardation of a vehicle. From now on we settle for talking about heavy braking, since it is obvious that this is an acceleration. These sensors have an update rate of 30-100 updates per second (30-100 Hertz) which is sufficient resolution for said problem. A large obstacle for using the built-in accelerometer in the mobile phone for said problem is that the mobile phone at normal operation and use continuously changes position in the car and that it is therefore complicated and computationally demanding to continuously calculate the angle between the sensitivity axes of the accelerometers in the mobile phone and the gravitational vector. It is also necessary that such a calculation has access to supplementary information, such as speed from GNSS receivers. To compensate the measurements from the accelerometers in the mobile phone for influence from the Earth's gravity thus requires a considerable computational power, which results in that the battery life is significantly shortened.
The invention relates to a method, device or program for calculating a high resolution acceleration signal from a low resolution measurement of speed. The invention further relates to a method, device or program for calculating a quality index associated with said acceleration signal.
The invention further relates to a method, device or program for detecting strong acceleration or retardation from said calculated acceleration signal and quality index.
These methods are achieved by parametric modelling of the dynamics of said speed measurement; means for calculating an acceleration from the parametric model; means for calculating a quality index for said calculated acceleration from calculated quality of said parametric model and said quality measure of said speed measurement.
The invention will be described in more detail in the following with reference to the attached drawings, which illustrate examples of selected embodiments, where:
Throughout the drawings the same reference numbers are used for similar or corresponding elements.
The proposed invention overcomes difficulties mentioned in the background by replacing the accelerometer (with update rate 30-100 Hertz) as a sensor by only a GNSS receiver (with update rate 1 Hertz), where the accelerometer's direct measurement of acceleration is replaced by an indirect measurement of acceleration through measured speed in combination with a parametric model or description of the motion. The challenge with this approach is multiple, including choice of parametric model and to reliably estimate the parameters in the parametric model in the presence of discontinuities and divergent values in measurement data. It is well known that speed data from a GNSS receiver contains isolated measurement points of poor quality, and periods of poor measurement data due to poor coverage.
It is not known to the inventors any electronic aid where an acceleration signal of high update rate is calculated from a speed signal with low update rate from a GNSS receiver, which at the same time ensures the validity of the calculated acceleration signal through a calculated quality index which depends on the quality of the original speed signal in combination with the quality of the parametric model that is used for describing the dynamics of the motion. Further, it is not known how such a quality index together with a calculated acceleration signal may be used to detect heavy braking of a vehicle during travel in a robust manner.
For each time tk the three data blocks times {tk−N, . . . , tk . . . , tk+N}, speed measurements (in the direction of the motion) {vk−N, . . . , vk . . . , vk+N}, and quality measure {qk−N, . . . , qk, . . . , qk+N} are saved, where vk and qk symbolize the speed and data quality provided by the GNSS receiver at time tk.
After step S1 the diagram is divided into two branches, step S2 and S4 respectively. It is obvious to the expert that since these different branches are independent from each other, the execution may also be done sequentially.
In step S2 a parametric motion model sk(θ, t) is adapted to the speed data collected in step S1. The parametric motion model sk(θ, t), which is unequivocally described by the parameters (the number of free parameters is L+1) θ={α0, α1, . . . , αL}, may be a linear function, non-linear function, discontinuous function that describes a relationship between times and parameters to speed. In a preferred embodiment of the invention the motion model is a polynomial of order L, where L=0, 1, 2, 3, . . . . In a preferred embodiment of the invention the motion model is a second order polynomial, i.e. sk(θ, t)=α0+α1(t−tk)+α2(t−tk)2, which exemplifies a linear function in the parameters θ={α0, α1, α2}. Thus in step S2 an adjustment of the parameters θ={α0α1, . . . , αL} is done so that the output signal from the model sk(θ, t) fits as closely as possible to collected speed data {vk−N, . . . , vk . . . , vk+N}, resulting in numeric values which are denoted {circumflex over (θ)}k where index k indicates that it is that parameter set which is applicable to the speed block centered around tk, {vk−N, . . . , vk . . . , vk+N}. Adaptation of the parameters in the motion model is done at each time tk based on surrounding data. Mathematically adaptation can be done by minimizing a cost function Vk(θ), i.e. {circumflex over (θ)}k=argminθVk(θ) where the cost function is a function of the difference between the measured value of the speed and the model's predicted speed as a function of the searched parameters, based on the measured values in the current block of data. The cost function may for example be a sum of squares of the errors, weighted sum of squares of the errors, maximum absolute value of the error or such that it maximizes the probability for the observed data (maximum likelihood) (which can be solved as a minimizing problem to fit into the framework of minimizing a cost function). To the skilled person it is obvious that measurement data in the cost function can be weighted with the quality measures {qk−N, . . . , qk, . . . , qk+N} to minimize the influence of measured values with high uncertainty in the model adaptation. In a proposed design of the invention a weighted sum of squares of the mathematical terms Vk(θ)=Σl=kk+N wl(vl−sk(θ, tl))2=Σl=k−Nk+N wl(v1−(α0+α1(tl−tk)+α2(tl−tk)2))2 where the second equality exemplifies the use of a second order polynomial, where the weights are suitable positive real numbers, for example forming a parabola where data close to the end points of the data block is weighted down for the benefit of a higher weight closer to the midpoint of the block. The solution is given in the example with a second order polynomial of the parameters {α0, α1, α2} which minimize the cost function.
In step S3 a residual or rest term is then calculated which describes the adaptation between model and measurement data. The residual is a scalar value which for example is given by the minimum value Vk({circumflex over (θ)}k) of the cost function or other above mentioned function of the error.
In step S4 a quality measure qkt, is calculated for data based on the sampling times {tk−N, . . . , tk, . . . , tk+N}. The mapping qkt←({tk−N, . . . , tk, . . . , tK+N}) can be done in several ways, for example by comparing the sampling intervals {tk+N−tk+N−1, . . . , tk−N+1−tk−N} with the nominal sampling period of GNSS receivers. At normal operational circumstances and at favorable receiving circumstances a GNSS receiver in a mobile phone typically has a sampling period tk−tk−1=1 second. If the sampling period of the GNSS receiver varies greatly it is an indicator that the GNSS receiver is having trouble calculating its position and speed, and measurement data is therefore typically of low quality. Examples of actual sampling periods for GNSS data during travel in a vehicle collected with an iPhone 5 is illustrated in
is used where typically T=1 second.
In step S5 a partial quality index δqktot is calculated for data at time tk by weighting together the residual (for example Vk({circumflex over (θ)}k), the quality measure qk of the GNSS receiver and the in the step S4 calculated quality measure qkt. It is obvious that these quality measures can be weighted together in several ways, where different weights are given to the different included quality measures. In a proposed embodiment we weight the quality measures together according to δqktot=β0Vk({circumflex over (θ)}k)+β1qk+β2qkt, where Vk({circumflex over (θ)}k) is a residual, and β0, β1, β2 are real valued weights which are positive, but not strictly positive.
In step S6 the acceleration âk is finally calculated at the time tk by differentiating the parametric model, i.e.
In a proposed embodiment with a motion model in the form of a second order polynomial is therefore {circumflex over (α)}k=α1.
In this embodiment the same time base is used for the resulting acceleration signal as for the original speed signal. To the skilled person it is obvious that the time base for the acceleration signal may be adjusted. The acceleration at an arbitrary time τ can be calculated according to
where k=argminl (abs(τ−tl)). In a proposed embodiment with a motion model in the form of a second order polynomial is therefore {circumflex over (α)}(τ)=α1+2α2(τ−tk) where k=argminl(abs(τ−tl)). Step S7 finishes the method.
The calculated acceleration at time tk is calculated from data block 410 through means 470. The calculated quality index qktot at time tk, on the other hand is calculated as the sum of the quality measures of 420, 410, and 430 and (the in the figure not depicted) intermediate blocks corresponding to data blocks centered around the times tk−2N+1 . . . tk−1 and tk+1 . . . tk+2N−1 first through means 460, 462, and 464 (and corresponding not depicted means 461 and 463 corresponding to the data blocks centered around the times tk−2N+1 . . . tk−1 and tk+1 . . . tk+2N−1) and in subsequent means 450. The means 460, 461, 462, 463 and 464, calculate partial quality indices {δqk−2Ntot, . . . , δqk+2Ntot}. Means 450 weights together the partial quality indices from said means 460, 461, 462, 463 and 464 to the final quality index qktot.
Weighting of the quality index qktot can be done in several ways. In a proposed design a direct summation is used, i.e. qktot=Σl=k−2Nk+2Nδqltot. Other ways of weighting together comprise a weighted sum where the weight for the different partial quality indices is determined for example by the distance from the center point.
440 illustrates the stream of output data from the proposed embodiment of the invention, i.e. comprising acceleration signal and associated quality index. As the time chart indicates the processing of data is block based. In the proposed design in
To the skilled person it is obvious that the built-in time delay, when needed, can be reduced by using a data block where tk is not centered in the block, for example by using only historic values.
Examples of test quantity comprise the ratio between the calculated acceleration and the calculated quality index. In a proposed embodiment of the embodiment is
where c is a strictly positive real constant. In a proposed embodiment of the invention 0<c<10 is used.
In step S11 the in step S10 calculated test quantity TEST QUANTITY is compared with a threshold value (THRESHOLD); the threshold value may be constant, time varying, or data dependent. In a proposed embodiment a constant threshold value is used. A time varying threshold may in one embodiment depend on time of day, where a higher threshold is allowed during the daylight hours, controlled through a clock. A data dependent threshold value may be linked to the measured speed, where an increased speed may imply a different threshold level (higher or lower) compared to a lower speed.
If TEST QUANTITY is lower or equal to THRESHOLD the method finishes in step S13. If the test quantity is larger than the threshold value a flag (FLAG) is set in step S12 indicating heavy braking. FLAG indicates that heavy braking has occurred. In a proposed embodiment the number of set flags during a drive is stored. In a proposed embodiment the total number of set flags during a premium period for a car insurance is set, or other time period linked to a car insurance. In a proposed embodiment the times when the flag was set are stored.
The method finishes in step S13.
An acceleration signal and quality index from a circuit diagram of an embodiment of signal processing in accordance with the invention thus enables more reliable detection of heavy braking of vehicles only using output data from a GNSS receiver, than when only the available speed signal from a GNSS receiver is used.
The present invention may be implemented as a microprocessor, a digital signal processor (DSP), or a combination with corresponding software. In a design the method may be implemented as a computer program which is installed in a mobile phone or computer via computer-readable media such as CD, DVD, USB memory, hard drive, via AppStore or Android Market, etc. The steps of the method are then executed in this program.
Another possible implementation is to use programmable logic in FPGA (field programmable gate arrays) or ASIC (application specific integrated circuit).
The above described embodiments should be regarded as examples of the present invention. The skilled person realizes that different modifications, combinations and changes of the described embodiments may be done without diverting from the scope of the present invention. The scope of the present invention is however defined by the enclosed patent claims.
| Number | Date | Country | Kind |
|---|---|---|---|
| 1330082-7 | Jun 2013 | SE | national |
| Filing Document | Filing Date | Country | Kind |
|---|---|---|---|
| PCT/SE2014/050791 | 6/26/2014 | WO | 00 |
