Method for predicting the behavior of a target vehicle

Abstract

The present invention relates to a method for predicting the behavior of a target vehicle (2) in a maneuver space, wherein a plurality of maneuvers that can be executed in the future are set for the target vehicle in the maneuver space, wherein the target vehicle (2) located in the maneuver space is observed by a plurality of observing vehicles (4a, 4b, 4c, 4d), wherein for each maneuver that can be executed in the future by the target vehicle (2), based on the observations made by the respective observing vehicles (4a, 4b, 4c, 4d), a single probability distribution for the maneuver that can be executed in the future is determined, and wherein for each maneuver that can be executed in the future by the target vehicle (2), a total probability distribution for the corresponding maneuver that can be executed in the future by the target vehicle (2) is determined from the plurality of single probability distributions.

Description

Technical Field

[0001] The present invention relates to a method and a system for predicting the behavior of a target vehicle. Background Art

[0002] In principle, it is possible that a vehicle located in road traffic is detected by sensors of another vehicle, and based on this, the behavior of the vehicle is analyzed.

[0003] Document EP 2 096 612 B1 describes a system and a computer program for creating a route control plan.

[0004] A method for controlling the light distribution of a vehicle's headlights is known from document EP 2 957 462 A1.

[0005] Document WO 2019 / 138485 A1 describes a method for predicting a collision.

[0006] Document US 2016 / 0357188 A1 describes measures for operating an intelligent vehicle, wherein a 3D model of the field of view of the vehicle sensors is generated. Additionally, information is received from nearby vehicles to compensate for blind spots in the field of view of the driver and sensors of the intelligent vehicle. Furthermore, traffic information is received, and based on the received traffic information and blind spot information, multiple 3D models are adjusted. Multiple 3D models are aggregated to generate a comprehensive 3D model combined with detailed map information. Summary of the Invention

[0007] Based on this background, the object of the present invention is to predict the behavior of a vehicle in road traffic.

[0008] This object is solved by a method and a system having the features of the independent patent claims. Embodiments of the method and the system are known from the dependent patent claims and the description.

[0009] The method according to the present invention is configured to predict the behavior of a target vehicle, for example configured as a motor vehicle, in a maneuver space such as in road traffic. Here, a plurality of maneuvers that can be performed in the future are set and / or considered for the target vehicle in the maneuver space, wherein the target vehicle located in the maneuver space is observed by a plurality of observing vehicles, for example by a plurality of other motor vehicles, and wherein for each maneuver that can be performed in the future for the target vehicle, based on the observations made by the respective observing vehicles, a single probability distribution for the respective maneuver that can be performed in the future is determined and / or obtained. Furthermore, for each maneuver that can be performed in the future for the target vehicle, a total probability distribution for the respective maneuver that can be performed in the future for the target vehicle is determined and / or obtained from a plurality of, in particular all, single probability distributions.

[0010] In the design, the target vehicle is observed by i observing vehicles, where m predictable and / or to-be-predicted maneuver actions are set for and / or considered for the target vehicle. For the respective n-th maneuver action, where 1 <= n <= m, the total probability distribution Pr(M = Mn|Observation_1,..., Observation_i) is determined based on the i observations.

[0011] Here, it is feasible that based on the j-th observation_j of the target vehicle, where 1 <= j <= i, the j-th total probability distribution Pr(M = Mn|Observation_j) for the n-th maneuver action is determined by the j-th observing vehicle and forwarded to the next j + 1-th observing vehicle. Considering the j + 1-th observation_j+1 of the target vehicle for the n-th maneuver action, the j + 1-th total probability distribution Pr(M = Mn|Observation_j, Observation_j+1) is determined by the next j + 1-th observing vehicle.

[0012] Therefore, it is feasible that the total probability distribution for the n-th maneuver action is determined iteratively through multiple, for example, i observation results of different observing vehicles. The more observations are performed by different observing vehicles, the more accurate the total probability distribution for the n-th maneuver action will be.

[0013] By definition, for the n-th maneuver action, the (j + 1)-th total probability distribution (j + 1) can be obtained by combining the j-th total probability distribution (j) with the (j + 1)-th single probability distribution (j + 1).

[0014] Supplementally, a hash value (Hashwert) can be provided and / or generated for each performed observation that has already been determined by the respective observing vehicle and considered for the total probability distribution, using which it is indicated that this observation has already been considered for the total probability distribution.

[0015] The target vehicle is observed by observing vehicles from different angles.

[0016] This method can be performed by at least one automatically or manually controlled observing vehicle, usually multiple automatically or manually controlled observing vehicles, for a manually or automatically controlled target vehicle.

[0017] In this method, the direction and / or speed of the target vehicle's travel can be considered as at least one maneuver action, for example, the change in the speed of the target vehicle, and thus the acceleration. For each maneuver action, forward travel, such as straight travel, left travel, right travel, and / or reverse travel, is considered, and for each maneuver action, maintaining a constant speed, accelerating, and / or braking are considered, and if necessary, standing still is also considered.

[0018] Furthermore, it is feasible to select, for example, m maneuvering actions set and / or taken into account or to be taken into account from a larger number of possible maneuvering actions, wherein the behavior of the target vehicle is predicted only based on the selected, for example, significant maneuvering actions.

[0019] The system according to the invention is configured to predict the behavior of a target vehicle in a maneuvering space and has sensors arranged in a plurality of observing vehicles and at least one computing unit arranged in at least one observing vehicle. Generally, at least one sensor for detecting the surroundings of the observing vehicle is arranged in each observing vehicle. In addition, each observing vehicle can include a computing unit. A plurality of maneuvering actions that can be executed in the future are set for the target vehicle in the maneuvering space, wherein at least one sensor of each observing vehicle is respectively configured to observe the target vehicle located in the maneuvering space. The at least one computing unit is configured to determine and / or obtain, for each maneuvering action that can be executed in the future of the target vehicle - based on the observations made by at least one sensor of the corresponding observing vehicle - a single probability distribution for the corresponding maneuvering action that can be executed in the future, and also to determine and / or obtain a total probability distribution for the corresponding maneuvering action that can be executed in the future of the target vehicle from a plurality of, in particular all, single probability distributions.

[0020] Each observing vehicle can have at least one sensor configured to receive electromagnetic waves. It is possible here that at least one sensor is configured as an optical sensor, for example as a camera. Alternatively or additionally, it is also possible that at least one sensor is configured as a radar sensor, an infrared sensor or a lidar sensor. Such a sensor can also be configured as an ultrasonic sensor. It is also feasible that the corresponding observing vehicle has a plurality of sensors for detecting the surroundings, wherein these sensors can be configured identically or differently. In a design of the proposed method with a design of the proposed system, it is now feasible to perform observations of the target vehicle from different perspectives by the sensors arranged on the observing vehicle. These observations performed separately by the sensors are evaluated by the at least one computing unit. It is feasible here to determine a single probability for the corresponding maneuvering action based on the respective observation results.

[0021] Furthermore, each observing vehicle participating in the method has a communication device for transmitting and receiving the observation results obtained by the sensors, the corresponding single probability distributions based on the respective observation results, and / or the iteratively provided total probability distribution.

[0022] Typically, each observing vehicle includes a computing unit configured to determine a single probability for a corresponding maneuver based on a corresponding observation of a target vehicle. Additionally, each computing unit can be configured to determine not only the corresponding single probability distribution but also the j-th total probability distribution taking into account the (j - 1)-th total probability distribution provided by one of the other observing vehicles and taking into account the j-th single probability distribution. If only a part of all observing vehicles have respective computing units for determining single probability distributions and / or total probability distributions, it is feasible to transmit individual observations from observing vehicles without a computing unit to the corresponding observing vehicles having a computing unit, where the computing unit is configured to determine the total probability distribution taking into account different observations from different observing vehicles.

[0023] Regardless of how the total probability distribution is calculated within the scope of this method based on different observations and / or single probabilities, it is generally feasible to perform cooperative maneuver prediction for a target vehicle by means of multiple observing vehicles. The total probability distribution, at least one single probability distribution, and / or at least one individual observation result are exchanged between vehicles via signals transmitted by electromagnetic waves. For radio-based exchange of signals, the C2X (car-to-everything) or V2X (vehicle-to-everything) function is used, which allows signal exchange between individual vehicles, here the observing vehicles, but also allows signal exchange between a vehicle and other devices.

[0024] In a design, it is feasible for the observing vehicles to drive automatically. Additionally, it is feasible for the observing vehicles to jointly predict the behavior of a target vehicle as an additional traffic participant better than a single vehicle can. Thus, the maneuver to be predicted or forecast is not determined by a single observing vehicle but by multiple, especially all, observing vehicles around the target vehicle. It should be taken into account here that, in order to determine the total probability distribution, each individual observing vehicle, using at least one of its sensors and / or its computing unit, provides at least one independent distorted / coarse observation for the maneuver to be predicted. Here, all observations of the observing vehicles can be distorted, where the distortions of the observations occur independently of each other. It is feasible here that the observations can be based on at least one acquisition, usually multiple acquisitions, performed by a sensor configured as a camera. Here, the acquisitions are made from different perspectives and / or angles by different sensors.

[0025] In the design solution, the respective observations and / or individual probability distributions of all observed vehicles can be collected and merged / integrated via V2X communication, such as V2V (vehicle-to-vehicle) communication, between the respective observed vehicles, wherein the prediction of the corresponding maneuver is improved. Additionally, it is feasible that such improved prediction or forecast of the corresponding maneuver is transmitted via V2X communication to other vehicles that are manually controlled and thus non-autonomous, and thus the total probability distribution for the corresponding maneuver is shared with the other manually controlled vehicles. Based on the total probability distribution, warning indications can then be provided in the respective vehicles, or other protective measures can be performed automatically, semi-automatically, or with driver assistance.

[0026] In the design solution, the proposed method can be performed similar to a belief propagation method. Here, it is feasible that at least one observed vehicle sends a notification or message to at least one other observed vehicle via V2X communication for the cooperative prediction of the maneuver of a target vehicle. The notification includes information about the maneuver space, such as information about the arrangement and distribution of the maneuver space in the local space and a list of all maneuvers that can be achieved, executed, or are possible within the maneuver space by the target vehicle. Additionally, for each possible maneuver in the maneuver space, each observed vehicle determines an individual probability distribution by itself, wherein each individual probability distribution at least relates to the observation when the target vehicle has been detected by the observed vehicle. Additionally, a hash value can also be determined for the observations that have been considered for calculating the total probability distribution.

[0027] In the field of autonomous driving or traffic, it is necessary for autonomous vehicles, especially autonomous observation vehicles, to also get along well in mixed traffic - in which there may also be manually controlled vehicles in motion. It is feasible within the scope of this method to distinguish autonomous vehicles (such as observation vehicles) and manually driven or controlled vehicles (such as target vehicles) by classifying the vehicles. It is feasible here to determine the total probability of maneuvering actions that can be correspondingly executed taking into account the observations of multiple, especially all, observation vehicles for each manually driven vehicle. In addition, it is proposed to use measures of belief propagation for near-optimal decoding of channel coding methods in the field of signal processing for radio systems. Here, the guesses about the results of random experiments between different participants are iteratively processed, and thus the guesses are improved. The proposed method can be based on belief propagation, where the cooperative prediction of the maneuvering actions of the target vehicle can be determined by multiple, especially all, observation vehicles. Belief propagation is described in the article "Low Density Parity Check Codes" by Rajesh Poddar (mid-term paper in ELE539B in spring 2007).

[0028] Further advantages and design options of the present invention result from the description and the drawings.

[0029] It should be understood that the features mentioned above and to be explained below can be used not only in the combinations respectively described, but also in other combinations or alone, without departing from the scope of the present invention. Brief Description of the Drawings

[0030] The present invention is schematically illustrated in the drawings in connection with embodiments and is described schematically and in detail with reference to the illustrations.

[0031] Figure 1 Schematic diagram showing an embodiment of the system according to the present invention when executing an embodiment of the method according to the present invention. Detailed Description of the Invention

[0032] Figure 1The figure shows a schematic illustration of a motor vehicle, which is configured here as the target vehicle 2, located in or moving within a maneuvering space, which in this case is road traffic that also has additional motor vehicles, and the additional vehicles are configured here as observation vehicles 4a, 4b, 4c, 4d. Here, each of the observation vehicles 4a, 4b, 4c, 4d has a plurality of sensors 6 for observing and / or detecting the surroundings of the respective observation vehicle 4a, 4b, 4c, 4d. In addition, each of the observation vehicles 4a, 4b, 4c, 4d has a computing unit 8 and a communication device 10, and only the antenna of the communication device is shown here, where the communication devices 10 are configured to perform V2X (German: Fahrzeug-zu-allem, English: vehicle-to-everything, Chinese: vehicle-to-everything) communication, at least V2V communication, between the respective observation vehicles 4a, 4b, 4c, 4d. It is proposed here that the provided sensors 6, computing units 8 and communication devices 10 distributed over the plurality of observation vehicles 4a, 4b, 4c, 4d can also be configured, by definition, as components of an embodiment of the system 10 according to the present invention.

[0033] In an embodiment of the method according to the present invention proposed here, the behavior of the target vehicle 2 in the maneuvering space is predicted. It is feasible here that the target vehicle 2 can perform different maneuvers within the maneuvering space, and these maneuvers can be defined, for example, by changing the direction of the target vehicle 2, such as by driving straight or turning left or right, and / or by changing the speed of the target vehicle 2, such as by accelerating or braking.

[0034] In an embodiment of the method, a first observation (Observation_1) of the target vehicle 2 is made from a first perspective by at least one sensor 6 of the first observation vehicle 4a at a certain point in time. In addition, a second observation (Observation_2) of the target vehicle 2 is made from a second perspective by at least one sensor 6 of the second observation vehicle 4b. Independently of this, a third observation (Observation_3) and a fourth observation (Observation_4) of the target vehicle 2 can also be made, respectively, by one of the two other observation vehicles 4c, 4d from their respective perspectives, and at least one sensor 6 of the respective observation vehicle 4c, 4d detects the target vehicle 2 from this perspective.

[0035] Furthermore, it is proposed to select and / or consider four different maneuvers M1, M2, M3, M4 from a large number of possible maneuvers M1, M2, M3, M4, M5, M6,... in the maneuver space for the target vehicle 2 if necessary, wherein for each of the selected maneuvers M1, M2, M3, M4, a cooperative total probability distribution is provided based on the observations (Observation_1, Observation_2, Observation_3, Observation_4) of all observing vehicles 4a, 4b, 4c, 4d, taking into account the different perspectives of the observing vehicles 4a, 4b, 4c, 4d relative to the target vehicle 2.

[0036] Here, a total of n = 4 maneuvers are to be considered, wherein a total of i = 4 observations are carried out for the target vehicle 2. Here, the j-th observation (Observation_j) is carried out by the j-th observing vehicle 4a, 4b, 4c, 4d, where 1 <= j <= i and i = 4. Here, the corresponding j-th observation is provided and / or detected by at least one sensor 6 of the j-th observing vehicle 4a, 4b, 4c, 4d. Based on an arbitrary j-th observation, the computing unit 8 of the corresponding j-th observing vehicle 4a, 4b, 4c, 4d determines and / or obtains, for example calculates, a single probability distribution for each of the total n = 4 maneuvers. Thus, based on the j-th observation of the j-th observing vehicle 4a, 4b, 4c, 4d, a single probability distribution Pr(M = Mn|Observation_j) is obtained and / or determined for the n-th maneuver respectively. Based on multiple Observation_j and / or single probability distributions Pr(M = Mn|Observation_j), the total probability distribution Pr(M = Mn|..., Observation_j,...) is determined from these single probability distributions. Furthermore, for each j-th observation, Observation_j also provides a hash value hash_Observation_j.

[0037] In an embodiment of the method proposed here, according to the definition of the total probability distribution or the single probability distribution Pr(M = Mn|Observation_1), for the n = 4 possible maneuvers related to the first Observation_1 of the first observing vehicle, the following information is transmitted from the first observing vehicle 4a to at least one of the other observing vehicles 4a, 4b, 4c, 4d:

[0038] The information sent by the first observing vehicle 4a is:

[0039] Maneuver space: M1, M2, M3, M4

[0040] Pr(M = M1|Observation_1);

[0041] Pr(M = M2|Observation_1);

[0042] Pr(M = M3|Observation_1);

[0043] Pr(M = M4|Observation_1);

[0044] hash_Observation_1;

[0045] If the second observation vehicle 4b obtains a message with this information from the first observation vehicle 4a and can provide additional, here the second Observation_2, then a new total probability distribution is calculated taking into account the previous first Observation_1 and its own new second Observation_2. Then the following information is sent:

[0046] Pr(M = M1|Observation_1, Observation_2);

[0047] Pr(M = M2|Observation_1, Observation_2);

[0048] Pr(M = M3|Observation_1, Observation_2);

[0049] Pr(M = M4|Observation_1, Observation_2);

[0050] hash_Observation_1, hash_Observation_2;

[0051] If the third observation vehicle 4c receives this information and can provide a third Observation_3, then a new total probability distribution for these four maneuver actions is calculated taking into account the previous Observation_1, Observation_2 and its own new third Observation_3. Then the following information is sent:

[0052] Pr(M = M1|Observation_1, Observation_2, Observation_3);

[0053] Pr(M = M2|Observation_1, Observation_2, Observation_3);

[0054] Pr(M = M3|Observation_1, Observation_2, Observation_3);

[0055] Pr(M = M4|Observation_1, Observation_2, Observation_3);

[0056] hash_Observation_1, hash_Observation_2, hash_Observation_3;

[0057] By repeatedly applying Bayes' theorem, the corresponding total probability distribution can be iteratively expanded for each of the four maneuver actions in total.

[0058] The fourth observation vehicle 4d that receives the information of the message of the third observation vehicle 4c can then, for example, incorporate its fourth Observation_4 into the total probability distribution.

[0059] Pr(M = M1|Observation_1, Observation_2, Observation_3, Observation_4);

[0060] Pr(M = M2 | Observation_1, Observation_2, Observation_3, Observation_4);

[0061] Pr(M = M3 | Observation_1, Observation_2, Observation_3, Observation_4);

[0062] Pr(M = M4 | Observation_1, Observation_2, Observation_3, Observation_4);

[0063] hash_Observation_1, hash_Observation_2, hash_Observation_3, hash_Observation_4;

[0064] The quality of the total probability distribution is improved by adding additional observations, thereby improving the quality of the prediction.

[0065] By means of the hash values hash_Observation_1, hash_Observation_2, hash_Observation_3, hash_Observation_4, it is ensured that an observation is not included in the calculation multiple times. In the example shown, it can thus be ensured that, for example, the participating third observation vehicle 4c is not counted again in Observation_1 and Observation_2.

[0066] List of reference numerals:

[0067] 2 Target vehicle

[0068] 4a, 4b, 4c, 4d Observation vehicles

[0069] 6 Sensor

[0070] 8 Calculation unit

[0071] 10 Communication device

[0072] 12 System

Claims

1. A method for predicting the behavior of a target vehicle (2) in a maneuver space, in which a plurality of maneuvers that can be executed in the future are set for the target vehicle in the maneuver space, where, A target vehicle (2) located in a maneuver space is observed by a plurality of observing vehicles (4a, 4b, 4c, 4d). For each maneuver that the target vehicle (2) can perform in the future, the corresponding observing vehicle (4a, 4b, 4c, 4d) performs the observation. It is characterized in that, based on the observations performed by each observing vehicle (4a, 4b, 4c, 4d), a single probability distribution of the corresponding maneuver that can be performed in the future is determined. For each maneuver that the target vehicle (2) can perform in the future, the total probability distribution of the corresponding maneuver that the target vehicle (2) can perform in the future is determined by a plurality of single probability distributions. Wherein, the target vehicle (2) is observed by i observing vehicles (4a, 4b, 4c, 4d). For the target vehicle (2), m maneuvers are set. For the corresponding nth maneuver, herein, 1 <= n <= m, the total probability distribution Pr(M = Mn|Observation_1,..., Observation_i) based on i observations is determined. Wherein, the single probability distributions of all observing vehicles (4a, 4b, 4c, 4d) are collected and combined via V2X communication between the respective observing vehicles (4a, 4b, 4c, 4d). The prediction of the corresponding maneuver is transmitted to other manually controlled vehicles via V2X communication. Thus, the total probability distribution of the corresponding maneuver is shared with other manually controlled vehicles. A warning indication is provided in the corresponding other manually controlled vehicles based on the total probability distribution.

2. The method according to claim 1, characterized in that, Based on the jth observation_j of the target vehicle (2), herein, 1 <= j <= i, the jth observing vehicle (4a, 4b, 4c, 4d) determines the jth single probability distribution Pr(M = Mn|Observation_j) for the nth maneuver and transmits it to the j + 1th observing vehicle (4a, 4b, 4c, 4d). Considering the j + 1th observation_j + 1 of the target vehicle (2), the j + 1th observing vehicle determines the j + 1th total probability distribution Pr(M = Mn|Observation_j, Observation_j + 1) for the nth maneuver.

3. The method according to claim 1 or 2, characterized in that, For each performed observation that has been determined by each observing vehicle (4a, 4b, 4c, 4d) and has been considered for the total probability distribution, a hash value is provided, and using the hash value to indicate that the observation has been considered for the total probability distribution.

4. The method according to claim 1 or 2, characterized in that, The target vehicle (2) is observed by the observing vehicles (4a, 4b, 4c, 4d) from different perspectives.

5. The method according to claim 1 or 2, wherein the method is performed by at least one automatically controlled observing vehicle (4a, 4b, 4c, 4d) for a manually controlled target vehicle (2).

6. The method according to claim 1 or 2, characterized in that, The driving direction of the target vehicle (2) and / or the speed of the target vehicle (2) are considered as at least one maneuver.

7. The method according to claim 1 or 2, characterized in that, The maneuvers to be considered are selected from a large number of possible maneuvers.

8. A system for predicting the behavior of a target vehicle (2) in a maneuver action space, wherein, The system (12) has sensors (6) arranged in a plurality of observing vehicles (4a, 4b, 4c, 4d) and at least one computing unit (8) arranged in at least one observing vehicle (4a, 4b, 4c, 4d), wherein a plurality of future-executable maneuver actions are set for a target vehicle (2) in a maneuver action space, wherein at least one sensor (6) of each observing vehicle (4a, 4b, 4c, 4d) is respectively configured to observe the target vehicle (2) located in the maneuver action space, and wherein the at least one sensor (6) of the respective observing vehicle (4a, 4b, 4c, 4d) performs an observation for each future-executable maneuver action. It is characterized in that the at least one computing unit (8) is configured to determine, for each future-executable maneuver action of the target vehicle (2), a single probability distribution for the respective future-executable maneuver action based on the observations performed by the at least one sensor (6) of the respective observing vehicle (4a, 4b, 4c, 4d), and to determine a total probability distribution for the respective future-executable maneuver action of the target vehicle (2) from the plurality of single probability distributions. Wherein the target vehicle (2) is observed by i observing vehicles (4a, 4b, 4c, 4d), wherein m maneuver actions are set for the target vehicle (2), and wherein, for the respective n-th maneuver action, here 1 <= n <= m, a total probability distribution Pr(M = Mn|observation_1,..., observation_i) based on i observations is determined. Wherein the individual probability distributions of all observing vehicles (4a, 4b, 4c, 4d) are collected and combined via V2X communication between the individual observing vehicles (4a, 4b, 4c, 4d), wherein the prediction of the respective maneuver action is transmitted via V2X communication to other vehicles under manual control, so that the total probability distribution for the respective maneuver action is shared with the other vehicles under manual control, and wherein a warning indication is provided in the respective other vehicles under manual control based on the total probability distribution.

Images