METHOD FOR PREDICTING A DRIVING MANEUVER IN A DRIVER ASSISTANCE SYSTEM

Abstract

A method for predicting a driving maneuver in a driver assistance system of a motor vehicle. Based on dynamic data of the vehicle detected with the aid of sensors, a history of a driving maneuver, which is not yet completed, is recorded and, based on the history, a future course of the driving maneuver is predicted. Multiple hypotheses for potentially occurring driving maneuvers are represented by respective associated sets of parameters. A theoretical history of the driving maneuver is calculated based on the parameters for each hypothesis. The recorded history is compared to at least one of the theoretical histories. Based on a similarity degree, the hypothesis which describes the presently occurring driving maneuver with the highest probability is determined and output as the prediction result.

Claims

1. A method for predicting a driving maneuver in a driver assistance system of a motor vehicle, the method comprising the following steps: recording, based on dynamic data of the vehicle detected using sensors, a history of a driving maneuver which is not yet completed; predicting, based on the history, a future course of the driving maneuver; representing multiple hypotheses for potentially occurring driving maneuvers by respective associated sets of parameters; calculating a theoretical history of the driving maneuver based on the parameters for each hypothesis of the hypotheses; comparing the recorded history to at least one of the theoretical histories; determining and outputting as a prediction result, based on a degree of similarity of the recorded history and the at least one of the theoretical histories, a hypothesis which describes the presently occurring driving maneuver with the highest probability.

2. The method as recited in claim 1, wherein, in addition to the prediction result, a quality value is output, which is calculated based on the degree of similarity and indicates a reliability of the prediction result.

3. The method as recited in claim 1, wherein the dynamic data of the vehicle detected using the sensors include at least a driving velocity and a yaw rate.

4. The method as recited in claim 3, wherein each history indicates a yaw angle of the vehicle as a function of a path covered by the vehicle.

5. The method as recited in claim 4, wherein the sets of parameters for each hypothesis encompass at least one maximum yaw rate which is reached during the driving maneuver.

6. The method as recited in claim 5, wherein the theoretical histories are calculated based on a model in which the yaw rate in a first phase of the driving maneuver increases linearly to a maximum yaw rate, remains constant in a second phase, and decreases linearly in a third phase.

7. The method as recited in claim 1, wherein the prediction result is a sum of the histories weighted according to degree of similarity for multiple hypotheses which were compared to the recorded history.

8. The method as recited in claim 1, wherein a selection of the hypotheses for which the comparison to the recorded history is carried out is dependent on additional information about the driving state of the vehicle and/or the traffic infrastructure.

9. A driver assistance system for a motor vehicle, comprising: a prediction module configured to predict a driving maneuver in a driver assistance system of a motor vehicle, the prediction module configured to: recording, based on dynamic data of the vehicle detected using sensors, a history of a driving maneuver which is not yet completed, predict, based on the history, a future course of the driving maneuver, represent multiple hypotheses for potentially occurring driving maneuvers by respective associated sets of parameters, calculate a theoretical history of the driving maneuver based on the parameters for each hypothesis of the hypotheses, compare the recorded history to at least one of the theoretical histories, determine and output as a prediction result, based on a degree of similarity of the recorded history and the at least one of the theoretical histories, a hypothesis which describes the presently occurring driving maneuver with the highest probability.

10. A non-transitory computer-readable storage medium on which is stored program code for predicting a driving maneuver in a driver assistance system of a motor vehicle, the program code, when executed by a computer of the driver assistance system, causing the computer to perform the following steps: recording, based on dynamic data of the vehicle detected using sensors, a history of a driving maneuver which is not yet completed; predicting, based on the history, a future course of the driving maneuver; representing multiple hypotheses for potentially occurring driving maneuvers by respective associated sets of parameters; calculating a theoretical history of the driving maneuver based on the parameters for each hypothesis of the hypotheses; comparing the recorded history to at least one of the theoretical histories; determining and outputting as a prediction result, based on a degree of similarity of the recorded history and the at least one of the theoretical histories, a hypothesis which describes the presently occurring driving maneuver with the highest probability.

Description

BRIEF DESCRIPTION OF THE DRAWINGS

[0019] FIG. 1 shows a block diagram of a prediction module with the aid of which the method according to an example embodiment of the present invention is executable.

[0020] FIG. 2 shows an illustration of a turning maneuver “turning left.”

[0021] FIG. 3 shows a graphical representation of a history for the turning maneuver according to FIG. 2.

[0022] FIG. 4 shows an illustration of a turning maneuver “turning right.”

[0023] FIG. 5 shows a graphical representation of a history for the turning maneuver according to FIG. 4.

[0024] FIG. 6 shows a graphical representation for explaining a test of a hypothesis for a driving maneuver.

[0025] FIG. 7 shows an illustration of possible turning maneuvers while turning into a multi-lane street.

DETAILED DESCRIPTION OF EXAMPLE EMBODIMENTS

[0026] FIG. 1 shows a prediction module 10 as a block diagram, which is part of a driver assistance system for motor vehicles. The prediction module is formed by suitable software which runs on a computer installed in the vehicle.

[0027] Hypotheses 14 for different driving maneuvers, for example for straight-ahead driving and different turning maneuvers, are stored in a hypotheses library 12. Each hypothesis 14 encompasses a set of parameters 16 which characterize the driving maneuver, as well as a history 18 calculated from these parameters, which indicates one or multiple variable(s) characterizing the state of the vehicle as a function of the time or, equivalent thereto, as a function of the path covered by the vehicle during the driving maneuver, for the duration of the driving maneuver. In particular, history 18 includes at least one function which indicates the curve of a dynamic variable of the vehicle, which may be measured using vehicle-internal sensors. The output signals of these sensors form input variables which are continuously evaluated in prediction module 10.

[0028] In the shown example, these dynamic variables are velocity v of the vehicle as well as yaw rate d[psi]/dt. The yaw rate may be calculated from the signals of wheel speed sensors, for example, which also form the basis for the measurement of driving velocity v. Selectively or additionally, the yaw rate may also be measured with the aid of an inertial sensor. In recording module 20, path s covered by the vehicle is calculated from velocity v by integration over the time, and present yaw angle [psi] is also calculated by integration of the yaw rate, and function [psi](s) is recorded for a certain time interval, for example for several seconds, and is continuously updated. The reference point for path s and yaw angle [psi](s) in the process are the present position of the vehicle as well as the present orientation. According to the definition, s=0 thus indicates the present position, and [psi]=0 indicates the present yaw angle of the vehicle. For the sake of simplicity, it shall be assumed as an example that the input variables are updated using a fixed sample rate 1/T. After each update of the data, path [delta]s which the vehicle has covered during sample period T is then calculated from velocity v, and change [delta][psi] of yaw angle is accordingly calculated based on the present yaw rate. All values of s and [psi] stored in the recording module are then decreased by [delta]s or [delta][psi]. Since recording module 20 has only a limited memory capacity, the respective oldest value pair is discarded. As a result, the recording module at any point in time includes a history 22 which indicates the curve of function [psi](s) during a past time interval of a fixed length. This history 22 is reported to a comparison module 24 and is continuously updated at the cycle of sample rate 1/T.

[0029] In comparison module 24, the recorded history 22 is continuously compared to at least one of history 18 stored in hypotheses library 12. In the normal case, during straight-ahead driving of the vehicle, the recorded history 22 is compared to a straight-ahead driving history which is formed by function [psi](s)=0 for all s. As long as the most recent value of the recorded history 22 deviates from 0 by less than a certain threshold value, comparison module 24 supplies the value “straight-ahead driving” as prediction result 26. In addition, the comparison module supplies a quality value Q, which indicates how reliable prediction result 26 is. Quality value Q may be calculated, for example, by calculating the mean square deviation of the recorded yaw angles [psi] from the theoretical value (0 in this example), standardizing it, and then subtracting it from 1, so that Q=1 means that the result is absolutely certain, and Q=0 means that the result is completely uncertain.

[0030] As soon as the most recent value in the recorded history deviates significantly from 0, depending on the situation at least one other of the stored hypotheses 14 is tested by comparing the recorded history 22 to history 18 belonging to this hypothesis 14. In the process, it is assumed that the first value deviating from 0 marks the starting point of the stored history 18. Prediction result “straight-ahead driving” is initially still maintained, however quality value Q decreases with an increasing number of yaw angles deviating from 0. At the same time, a new quality value Q is formed for each tested hypothesis. During the standardization of this quality value, the statistical uncertainty due to the initially still small and then slowly expanding database is also taken into consideration. When the number of yaw angles deviating from 0 has reached a certain minimum value or, alternatively, when the quality value for the tested hypothesis has reached a certain minimum amount, this hypothesis is output as the new prediction result 26, and the output quality value Q is replaced with the quality value for this hypothesis.

[0031] In general, each of the stored hypothesis 18 may be tested as soon as the first yaw angle deviates significantly from 0. The selection of the tested hypotheses, however, may also depend on different factors which describe the present traffic situation. For example, no hypotheses for turning maneuvers need to be tested as long as velocity v is so high that no turning maneuver may take place. A test of turning hypotheses may also be dispensed with when the position of the vehicle located on a digital map shows that no turning possibility exists.

[0032] FIG. 2 illustrates a situation in which a vehicle 28 equipped with the driver assistance system has entered an intersection 30. In addition to hypothesis “straight-ahead driving,” hypotheses “turning left” and “turning right” would also have to be tested here. However, the yaw angle has already increased slightly compared to straight-ahead driving, so that hypothesis “turning left” appears to be more likely and is further tested. Based on the recorded history 22, it is possible to back-calculate the point at which history 18 for the turning maneuver has its beginning. The corresponding position 32 of the vehicle is shown with dotted lines in FIG. 2. The recorded history 22 is then calibrated in such a way that position 32 coincides with the starting point of history 18. The path of the vehicle covered since this point is indicated by an arrow 34 in FIG. 2.

[0033] FIG. 3 shows the recorded history 22 for the situation shown in FIG. 2 in a solid line. Theoretical history 18, which at the same time represents the prediction for the further course of the driving maneuver, is shown with dotted lines in FIG. 3. The associated presumable trajectory of the vehicle is indicated by an arrow 36 in FIG. 2.

[0034] Three phases may be distinguished in history 18 and the associated driving maneuver. In an introductory phase, which in FIG. 3 extends approximately from s=−12 to s=−5, the yaw rate increases linearly so that yaw angle [psi] grows quadratically. At s=−5, the yaw rate has reached a maximum value, which is then constantly maintained in the main phase of the maneuver up to approximately s=10. Thereafter, an expansion phase begins during which the yaw rate decreases linearly to the value 0. At the end of the driving maneuver, the yaw angle has increased by 90°, from −30° to +60°, corresponding to the hypothesis that a 90° arc is passed through during the driving maneuver.

[0035] For comparison, FIGS. 4 and 5 illustrate a situation in which a driving maneuver “turning right” takes place. Since the yaw angle decreases in a right-hand curve, histories 22 and 18 form a downward-sloping curve in FIG. 5. In addition, the radius of curvature of the trajectory in the main phase of the turning process is smaller when turning right than when turning left in countries with right-hand traffic. Accordingly, the yaw rate is larger in the main phase so that the curve in FIG. 5 is steeper and is compressed along the s axis. During the calculation of histories 18 for the different hypotheses, a standard width of the mutually intersecting streets may be assumed. When information about the actual dimensions of the intersection is known from a digital map, the calculation of the theoretical history may presently also take place based on these data.

[0036] FIG. 6 shows the test of a hypothesis for a maneuver “turning left” in a diagram similarly to FIGS. 3 and 5. The recorded history 22 is represented by individual measuring points here, which are shown as black dots in the diagram. The corresponding measuring points which represent the theoretical history 18 are shown as white dots. A similarity degree is now calculated for the test of the hypothesis, which indicates the degree of agreement between histories 18 and 22, for example based on the mean square deviation, as described above for quality value Q.

[0037] The determination of the prediction result may now take place, for example, by testing multiple possible hypotheses, and then selecting the hypothesis for which the degree of similarity is the greatest. When history 18 correctly describes the actual driving maneuver, the degree of similarity will converge toward the value 1 with increasing duration of the recording of history 22.

[0038] In general, however, none of the predefined hypotheses 14 will exactly match the recorded history 22. In this case, the degree of similarity remains smaller than 1 for all histories. It is now possible to select either the hypothesis having the greatest degree of similarity as the prediction result, or a new hypothesis is generated by forming a sum from the predefined hypotheses 18 which is weighted according to the degree of similarity.

[0039] FIG. 7 illustrates an example in which the driving maneuver is a turning process to the right into a two-lane intersecting street 38 (a one-way street). The hypotheses to be tested here are that vehicle 28 turns either into the right lane or the left lane of intersecting street 38.