Method for Judging Pedestrian Intention and Generating Trajectory under Vehicle-Pedestrian Interaction Based on Group Characteristics

Abstract

The present invention discloses a method for pedestrian intention judgment and trajectory generation under vehicle - pedestrian interaction based on group characteristics, belonging to the field of autonomous driving technology, including: acquiring the vehicle - pedestrian characteristics of the vehicle - pedestrian interaction scene, calculating the spatio - temporal proximity and time - to - collision through the vehicle - pedestrian characteristics and presetting thresholds to identify the pedestrians and vehicles with interaction; introducing an adjustable distance threshold as a constraint to construct the pedestrian group and vehicle group that affect the pedestrian interaction behavior; constructing a deep learning network and extracting the group characteristics of the pedestrian group and vehicle group respectively; according to the group characteristics, simulating the game interaction behavior between the pedestrian group and the vehicle group, fusing the group characteristics and the game characteristics between the groups to obtain the group fusion characteristics; according to the group fusion characteristics, performing pedestrian intention judgment and trajectory generation under vehicle - pedestrian interaction through a deep learning framework. The present invention can accurately describe pedestrian intentions and generate simulated trajectories, which can better assist in planning driving routes and improve traffic safety.

Description

Technical Field

[0001] The present invention belongs to the technical field of autonomous driving, and particularly relates to a method for judging pedestrian intention and generating trajectory under vehicle - pedestrian interaction based on group characteristics. Background Art

[0002] With the rapid development of artificial intelligence, sensor technology, and vehicle - to - everything (V2X) communication, autonomous driving has become a key innovative technology in intelligent transportation systems. Its core goal is to promote the intelligentization and sustainable development of transportation systems by reducing traffic congestion, improving road safety, and reducing carbon emissions. In recent years, significant progress has been made in autonomous driving technology in aspects such as perception systems, decision - making algorithms, and vehicle control, and some autonomous driving functions have been commercially applied.

[0003] Despite the significant progress of autonomous driving technology, there are still many challenges in practical applications. First, the safe interaction between autonomous vehicles and pedestrians is a weak link in current technology. The uncertainty and flexibility of pedestrian behavior are relatively high, and there is a lack of effective interaction mechanisms between autonomous vehicles and pedestrians, resulting in difficult accurate prediction of pedestrian intentions. Second, most existing research focuses on the behavior modeling of individual pedestrians or vehicles, ignoring the impact of group behavior on the interaction process, which makes the generated intention judgment and trajectory prediction less comprehensive. In addition, due to the lack of full consideration of complex real - world scenarios, mechanism models usually have problems with insufficient accuracy. Although learning models perform better in simulating pedestrian behavior, they still cannot fully cover the complex scenarios in vehicle - pedestrian interaction. These deficiencies limit the application of autonomous driving technology in complex traffic environments and reduce the public's trust in autonomous vehicles. Summary of the Invention

[0004] To solve the above - mentioned technical problems, the present invention proposes a method for judging pedestrian intention and generating trajectory under vehicle - pedestrian interaction based on group characteristics to solve the problems existing in the above - mentioned prior art.

[0005] To achieve the above object, in a first aspect, the present invention provides a method for judging pedestrian intention and generating trajectory under vehicle - pedestrian interaction based on group characteristics, including:

[0006] Obtain the vehicle - pedestrian characteristics of the vehicle - pedestrian interaction scenario, calculate the spatio - temporal proximity and time - to - collision through the vehicle - pedestrian characteristics, and identify the pedestrians and vehicles with interaction through the preset thresholds of spatio - temporal proximity and time - to - collision;

[0007] Introduce an adjustable distance threshold as a constraint to construct a pedestrian group and a vehicle group that affect pedestrian interaction behavior; construct a deep learning network, and extract the group characteristics of the pedestrian group and the vehicle group respectively;

[0008] According to the group characteristics, simulate the game interaction behavior between the pedestrian group and the vehicle group, fuse the group characteristics and the game characteristics between the groups to obtain the group fusion characteristics;

[0009] According to the group fusion characteristics, perform pedestrian intention judgment and trajectory generation in the vehicle-pedestrian interaction through a deep learning framework.

[0010] Preferably, identifying pedestrians and vehicles with interactions includes:

[0011] Obtain the vehicle-pedestrian characteristics of the vehicle-pedestrian interaction scenario; the vehicle-pedestrian characteristics include pedestrian position characteristics, vehicle position characteristics, pedestrian speed characteristics, and vehicle speed characteristics;

[0012] Calculate the spatio-temporal proximity through the vehicle-pedestrian characteristics and record it in a symmetric spatio-temporal proximity matrix;

[0013] Calculate the time to collision through the vehicle-pedestrian characteristics and record it in a symmetric relative collision time matrix;

[0014] Based on the spatio-temporal proximity matrix and the relative collision time matrix, set adjustable thresholds respectively, and identify pedestrians and vehicles with interactions according to the thresholds.

[0015] Preferably, extracting the group characteristics of the pedestrian group and the vehicle group respectively includes:

[0016] Introduce an adjustable distance threshold as a constraint, and respectively screen out the surrounding pedestrians and surrounding vehicles that have an impact on the behavior of interacting pedestrians;

[0017] Based on the surrounding pedestrians and the surrounding vehicles, introduce interacting pedestrians to construct a pedestrian group and a vehicle group;

[0018] Respectively construct a scene-adaptive deep learning network, and extract the group characteristics of the pedestrian group and the vehicle group through the deep learning network.

[0019] Preferably, the group characteristics include the internal state relationship of the group and the group position relationship;

[0020] The calculation formula for the internal state relationship of the group is:

[0021] ;

[0022] Where and are the state characteristics of each subject inside the pedestrian group or the vehicle group, and are learnable parameters, is the dimension of the state characteristics;

[0023] The calculation formula for the group position relationship is as follows:

[0024] ;

[0025] where, is the spatial distance between each entity within the pedestrian group or vehicle group; is an adjustable spatial threshold; is an indicator function.

[0026] Preferably, obtaining the group fusion feature includes:

[0027] According to the group features, construct a game strategy set for the pedestrian group and the vehicle group; through the game strategy set, establish the game utility of the pedestrian group and the vehicle group;

[0028] Based on the game utility, simulate the game interaction behavior of the pedestrian group and the vehicle group under the interaction between pedestrians and vehicles, and obtain the expected utility of the pedestrian group and the vehicle group;

[0029] Based on the expected utility, fuse the group features of the pedestrian group and the vehicle group to obtain the group fusion feature.

[0030] Preferably, the game utility includes the direct risk of collision, indirect risk, avoidance loss, and group utility; the indirect risk includes the minimum future relative distance, minimum future relative time, and conflict speed of larger participants.

[0031] Preferably, the game utility of the pedestrian group and the vehicle group is expressed as:

[0032] ;

[0033] where, is the utility of the pedestrian group or vehicle group under the strategy combination, is the minimum future relative distance of the pedestrian group or vehicle group under the strategy combination, is the minimum future relative time of the pedestrian group or vehicle group under the strategy combination, is the conflict speed of larger participants of the pedestrian group or vehicle group under the strategy combination.

[0034] Preferably, the process of pedestrian intention judgment and trajectory generation under the interaction between pedestrians and vehicles through the deep learning framework includes:

[0035] Take the group fusion feature as the input, construct a behavior intention feature extraction module under the deep learning framework, and extract the behavior intention features in the pedestrian-vehicle interaction scenario;

[0036] Perform shaping and scaling on the behavioral intention features to obtain the reward function of the interacting pedestrians under the vehicle-pedestrian interaction;

[0037] Use the reward function as the evaluation criterion to obtain the behavioral strategy of the interacting pedestrians under the vehicle-pedestrian interaction under the influence of the pedestrian group and the vehicle group;

[0038] Iteratively update the behavioral intention feature extraction module and the behavioral strategy to perform pedestrian intention judgment in heterogeneous interaction scenarios and generate pedestrian trajectories.

[0039] In a second aspect, the present invention also discloses a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, the steps of the method described in the first aspect are implemented.

[0040] In a third aspect, the present invention also discloses a computer program product, including a computer program, and when the computer program is executed by a processor, the steps of the method described in the first aspect are implemented.

[0041] Compared with the prior art, the present invention has the following advantages and technical effects:

[0042] The present invention provides a method for pedestrian intention judgment and trajectory generation under vehicle-pedestrian interaction based on group characteristics. First, obtain the vehicle-pedestrian characteristics of the vehicle-pedestrian interaction scenario, calculate the spatio-temporal proximity and the collision time through the vehicle-pedestrian characteristics, and identify the pedestrians and vehicles with interactions through the preset thresholds of the spatio-temporal proximity and the collision time; Secondly, introduce an adjustable distance threshold as a constraint to construct a pedestrian group and a vehicle group that affect the pedestrian interaction behavior; construct a deep learning network, and extract the group characteristics of the pedestrian group and the vehicle group respectively; Further, according to the group characteristics, simulate the game interaction behavior between the pedestrian group and the vehicle group, fuse the group characteristics and the game characteristics between the groups to obtain the group fusion characteristics; Finally, according to the group fusion characteristics, perform pedestrian intention judgment and trajectory generation under vehicle-pedestrian interaction through a deep learning framework.

[0043] The present invention accurately identifies the pedestrians and vehicles with interactions in the vehicle-pedestrian interaction scenario through the threshold indicators of spatio-temporal proximity and collision time; by constructing a pedestrian group and a vehicle group, a scene-adaptive deep learning framework is established, so as to accurately extract the group characteristics of the pedestrian group and the vehicle group respectively. Subsequently, introduce the game interaction behavior between the pedestrian group and the vehicle group, deeply fuse the group characteristics of each group, and finally use the deep learning framework to capture the behavioral intention and the optimal strategy of the pedestrians under the vehicle-pedestrian interaction in the fusion characteristics, so as to accurately judge the behavioral intention and generate high-precision pedestrian trajectories.

[0044] The present invention can effectively handle the influence of surrounding pedestrians and vehicles on pedestrian behavior in the interaction between pedestrians and vehicles, accurately describe pedestrian intentions and generate simulated trajectories, and can better assist in planning driving routes and improving traffic safety. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] The drawings constituting a part of this application are used to provide a further understanding of this application. The schematic embodiments of this application and their descriptions are used to explain this application and do not constitute an improper limitation to this application. In the drawings:

[0046] Figure 1 is a flowchart of the method according to an embodiment of the present invention;

[0047] Figure 2 is a schematic diagram of the indirect risk of the interaction between pedestrians and vehicles according to an embodiment of the present invention;

[0048] Figure 3 is a schematic diagram of the analysis of the reward function of the interacting pedestrians according to an embodiment of the present invention;

[0049] Figure 4 is a schematic diagram of the generated trajectory of the interacting pedestrians according to an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0050] It should be noted that, without conflict, the embodiments in this application and the features in the embodiments can be combined with each other. The following will refer to the drawings and combine the embodiments to detail this application.

[0051] It should be noted that the steps shown in the flowchart of the drawings can be executed in a computer system such as a set of computer-executable instructions, and although the logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in a different order than here.

[0052] Embodiment 1

[0053] As Figure 1 shown, in this embodiment, a method for judging pedestrian intentions and generating trajectories in the interaction between pedestrians and vehicles based on group characteristics is provided, including:

[0054] S1. Extract the pedestrian-vehicle characteristics from the pedestrian-vehicle interaction scenario, calculate the spatio-temporal proximity and the time to collision through the pedestrian-vehicle characteristics, and identify the pedestrians and vehicles with interaction through the preset thresholds of the spatio-temporal proximity and the time to collision; wherein, the pedestrian-vehicle characteristics include pedestrian position characteristics, vehicle position characteristics, pedestrian speed characteristics, and vehicle speed characteristics.

[0055] Specifically, S1 includes:

[0056] S11. Extract features such as pedestrian position features, vehicle position features, pedestrian speed features, and vehicle speed features from the vehicle - pedestrian interaction scenario. Calculate pairwise spatio - temporal proximity metrics based on the obtained features of pedestrian i and vehicle j, and record the pairwise spatio - temporal proximities in a symmetric proximity matrix.

[0057] The spatio - temporal proximity metric can be the Euclidean distance maintained by the pedestrian and the vehicle over a period of time.

[0058] S12. Calculate the pairwise relative collision time based on the obtained features of pedestrian i and vehicle j, and record the pairwise relative collision times in a symmetric pairwise relative collision time matrix.

[0059] The relative collision time can be the absolute time difference when pedestrian i and vehicle j pass through their trajectory intersection point (if any), and is defined as follows:

[0060] ;

[0061] In the formula, is the relative collision time, is the moment when pedestrian i passes through the intersection point of the trajectory with vehicle j, is the moment when vehicle j passes through the intersection point of the trajectory with pedestrian i.

[0062] S13. Based on the spatio - temporal proximity matrix and the relative collision time matrix, set adjustable thresholds to identify pairwise interacting pedestrians and vehicles, and record the interaction categories and the numbers of the pedestrians and vehicles involved in the interaction.

[0063] The adjustable threshold can include two aspects. On the one hand, it is the threshold for the spatio - temporal proximity set in S11. That is, the spatial distance (such as 10 meters) that pedestrian i and vehicle j should maintain within a certain time (such as 2 seconds) can be set. The spatial distance can be judged and calculated based on the braking performance of the vehicle and the speed limit during vehicle operation. On the other hand, a threshold is set for the relative collision time. A scenario with a lower is considered a more dangerous scenario. In actual application, 3 seconds can be taken as the threshold.

[0064] S2. Introduce an adjustable distance threshold as a constraint to construct groups of pedestrians and vehicles that affect pedestrian interaction behavior, construct a deep - learning network, and extract the group features of the pedestrian group and the vehicle group respectively.

[0065] Specifically, S2 includes:

[0066] S21. Calculate the pairwise distance between pedestrian i in the scenario and the interacting pedestrians, introduce an adjustable distance threshold under pedestrian interaction, screen the surrounding pedestrians that affect the behavior of the interacting pedestrians, and record them;

[0067] The paired distance may be the Euclidean distance between pedestrian i and the interacting pedestrian, and the adjustable distance threshold may be the intermediate value between the intimate distance and the social distance in social relation theory. In actual application, 4m may be taken.

[0068] S22. Calculate the paired distance between vehicle i in the scenario and the interacting pedestrian, introduce an adjustable distance threshold under vehicle-pedestrian interaction, screen the surrounding vehicles that have an impact on the behavior of the interacting pedestrian, and record them.

[0069] The paired distance may be the Euclidean distance between vehicle i and the interacting pedestrian, and the adjustable distance threshold may be comprehensively determined according to the speed limit of the road section, the traffic flow volume, and the braking performance of the vehicle. In actual application, 10m may be taken.

[0070] S23. As an innovative implementation method, based on the surrounding pedestrians and the surrounding vehicles, introduce the interacting pedestrian to construct a pedestrian group and a vehicle group, and respectively construct a scene-adaptive deep learning network framework to extract the group characteristics of the pedestrian group and the vehicle group.

[0071] The pedestrian group should include the interacting pedestrian under vehicle-pedestrian interaction and all the surrounding pedestrians that meet the pedestrian interaction adjustable distance threshold obtained by judgment.

[0072] The vehicle group should include the interacting pedestrian under vehicle-pedestrian interaction, the interacting vehicle, and all the surrounding vehicles that meet the vehicle-pedestrian interaction adjustable distance threshold obtained by judgment.

[0073] Due to different dynamic characteristics, appearances, etc., there should be different interaction modes within the pedestrian group and within the vehicle group. Based on this, in this embodiment, deep learning network frameworks are respectively constructed for the pedestrian group and the vehicle group to extract the group characteristics of the pedestrian group and the vehicle group in the vehicle-pedestrian interaction scenario.

[0074] Specifically, for each group, first calculate the internal state relationship of the group and the group position relationship .

[0075] The internal state relationship of the group can be calculated by means of dot product, embedded dot product, relational graph, etc. In this embodiment, the embedded dot product is adopted to calculate the internal state relationship of the group. Specifically as follows:

[0076] ;

[0077] In the formula, and is the state characteristic of each subject within the pedestrian group or vehicle group, and are learnable parameters, is the dimension of the state characteristic.

[0078] The internal positional relationship within the group can be calculated through methods such as position masks, embedded dot products, dot products, etc. In this embodiment, the position mask is used to calculate the internal positional relationship within the group, specifically as follows:

[0079] ;

[0080] In the formula, is the spatial distance between each subject within the pedestrian group or vehicle group. In practical applications, the Euclidean distance can be used; is an adjustable spatial threshold; is an indicator function, which takes 1 if the condition is satisfied, otherwise 0.

[0081] Through the state characteristic and the position characteristic, the internal influence relationship within the pedestrian group or vehicle group can be further calculated. It can be calculated through methods such as dot products, exponential function dot products, embedded dot products, etc. In this embodiment, the form of exponential function dot product combined with the softmax function is used for illustration, specifically as follows:

[0082] ;

[0083] Based on the internal influence relationship within the pedestrian group or vehicle group, methods such as graph convolution, multi-layer perceptron, convolution, etc. can be further used to extract its features. In this embodiment, graph convolution is used to aggregate its internal features, specifically as follows:

[0084] ;

[0085] In the formula, is the feature representation of the l-th layer of the pedestrian group or vehicle group, is a learnable parameter matrix.

[0086] S3. According to the group characteristics, simulate the game interaction behavior between the pedestrian group and the vehicle group, and fuse the group characteristics and the game characteristics between the groups to obtain the group fusion characteristics.

[0087] Specifically, S3 includes:

[0088] S31. As an innovative implementation method, construct the game strategy sets of the pedestrian group and the vehicle group, and establish the game utility of all strategy combinations of the pedestrian group and the vehicle group based on factors such as the speed characteristics and group sizes of the pedestrian group and the vehicle group.

[0089] The game strategy set may include deceleration, deflection, acceleration, etc. In this embodiment, the strategies for the pedestrian group and the vehicle group are adopted as (maintaining the original state) and giving way (deflection and deceleration).

[0090] The game utility may include parts such as the direct risk of collision, indirect risk, avoidance loss, and group utility. In this embodiment, the indirect risks in three aspects, namely the minimum future relative distance, the minimum future relative time, and the conflict speed of the larger participant, are selected to describe the game interaction behavior of the pedestrian group and the vehicle group. The schematic diagrams of the indirect risks in these three aspects are as shown in Figure 2 shown.

[0091] The minimum future relative distance evaluates the distance of another participant from the potential conflict point when a pedestrian or vehicle reaches the potential conflict point. The minimum future relative time evaluates the remaining time for another participant to reach the potential conflict point when a pedestrian or vehicle reaches the potential conflict point. The conflict speed of the larger participant evaluates the current speed of the vehicle when a pedestrian or vehicle reaches the potential conflict point.

[0092] When the pedestrian group and the vehicle group adopt different strategies, the speed, orientation, and the position of the potential conflict point will all change. Therefore, under different strategy combinations of the pedestrian group and the vehicle group, the game utility will change. The game utility of the pedestrian group and the vehicle group can be expressed as follows:

[0093] ;

[0094] In the formula, is the utility of the pedestrian group or the vehicle group under the strategy combination, is the minimum future relative distance of the pedestrian group or the vehicle group under the strategy combination, is the minimum future relative time of the pedestrian group or the vehicle group under the strategy combination, is the conflict speed of the larger participant of the pedestrian group or the vehicle group under the strategy combination.

[0095] S32. Based on the game utility, simulate the game interaction behavior of the pedestrian group and the vehicle group in the interaction between pedestrians and vehicles, and obtain the expected utilities of the pedestrian group and the vehicle group.

[0096] Based on the game utility, further calculate the expected utilities obtained after the pedestrian group and the vehicle group adopt a certain strategy, as follows:

[0097] ;

[0098] ;

[0099] ;

[0100] ;

[0101] Wherein, and are the expected utilities of the pedestrian group adopting the passing and yielding game strategies respectively. The same applies to the vehicle group. and are the probabilities of the pedestrian group and the vehicle group adopting the passing strategy.

[0102] According to the expected utilities of the pedestrian group and the vehicle group adopting the passing or yielding strategy, the calculation methods of the expected utilities of the pedestrian group and the vehicle group in this state under the vehicle-pedestrian interaction are as follows:

[0103] ;

[0104] ;

[0105] Wherein, and are the expected utilities of the pedestrian group and the vehicle group under the vehicle-pedestrian interaction.

[0106] According to the said expected utilities, the behaviors of the pedestrian group and the vehicle group can be simulated by means of Nash game, evolutionary game, leader-follower game, etc. In this embodiment, the evolutionary game is adopted to simulate the behaviors of the pedestrian group and the vehicle group. The replicator dynamic equations of the evolutionary game are as follows:

[0107] ;

[0108] ;

[0109] Wherein, and are the replicator dynamic equations of the pedestrian group and the vehicle group respectively, and the game behaviors of the pedestrian group and the vehicle group can be simulated by combining the strategy distributions and expected utilities of the pedestrian group and the vehicle group in the current state.

[0110] S33. Based on the said expected utilities, fuse the group characteristics of the pedestrian group and the vehicle group to obtain the group fusion characteristics.

[0111] After the evolutionary game models of the pedestrian group and the vehicle group are calibrated by the genetic algorithm, the corresponding utility equations are respectively:

[0112] ;

[0113] ;

[0114] After calculating the expected utility based on the utilities of the pedestrian group and the vehicle group under each strategy combination, it is fused with the group characteristics of the pedestrian group and the vehicle group, which can be carried out through dot product, cross product, embedded dot product, etc. In this embodiment, the dot product form is adopted, and the specific process is as follows:

[0115] ;

[0116] In the formula, F is the group fusion feature.

[0117] S4. According to the group fusion feature, perform pedestrian intention judgment and trajectory generation in the vehicle-pedestrian interaction through a deep learning framework.

[0118] Specifically, S4 includes:

[0119] S41. As an innovative implementation method, use the group fusion feature as the input to construct a behavior intention feature extraction module of the deep learning framework to extract the behavior intention features in the vehicle-pedestrian interaction scenario.

[0120] The deep learning framework can adopt network architectures such as multi-layer perceptron and convolutional network. In this embodiment, a multi-layer perceptron is used to extract the group fusion feature to obtain the behavior intention features of the interacting pedestrians, and the specific process is as follows:

[0121] ;

[0122] In the formula, is the reward value of the interacting pedestrians after being processed by the deep learning framework, which can be used to preliminarily judge the pedestrian intention.

[0123] S42. As an innovative implementation method, establish a deep learning algorithm module to reshape and scale the behavior intention features to obtain the reward function of the interacting pedestrians in the vehicle-pedestrian interaction.

[0124] Based on the reward value of the interacting pedestrians after being processed by the deep learning framework, it can be further reshaped and calculated using a deep learning network framework. Here, a generative adversarial learning framework can be used for the calculation to obtain the reward function of the interacting pedestrians in the vehicle-pedestrian interaction. The specific process is as follows:

[0125] ;

[0126] ;

[0127] ;

[0128] In the formula, is the reshaping term to help adjust the reward value, is the discount factor, A discriminator for determining whether the current trajectory is a real trajectory. The probability of the policy for taking action a in state s during training. The reward function after shaping and scaling.

[0129] S43. Using the reward function as the evaluation criterion, obtain the behavior strategy of the interacting pedestrians under the influence of the pedestrian group and the vehicle group in the vehicle-pedestrian interaction.

[0130] Using the reward function as the evaluation criterion, reinforcement learning algorithms such as Q-network, deep Q-network, soft actor-critic, etc. can be used to maximize the reward function to obtain the optimal behavior strategy of the interacting pedestrians under the influence of the pedestrian group and the vehicle group in the vehicle-pedestrian interaction.

[0131] S44. Iteratively update the behavior intention feature extraction module and the behavior strategy to perform pedestrian intention judgment and generate pedestrian trajectories in heterogeneous interaction scenarios.

[0132] The pedestrian intention judgment and trajectory generation methods are both dynamic processes, which focus on judging the behavior intention according to the current time t and generating the pedestrian trajectory at time t + 1.

[0133] The reward function and the behavior strategy are a process of feedback optimization. After the algorithm converges, the analysis of the reward function and the generation of pedestrian trajectories can be carried out, as shown in Figure 3 and Figure 4 respectively.

[0134] Advantages of this embodiment:

[0135] This embodiment uses adjustable spatio-temporal proximity and collision time threshold indicators to accurately identify the pedestrians and vehicles in interaction in the vehicle-pedestrian interaction scenario. Further, combined with the heterogeneous adjustable spatial distance indicator, it identifies the surrounding pedestrians and vehicles that have an impact on the interacting pedestrians, and constructs a pedestrian group and a vehicle group based on this, establishing a scene-adaptive deep learning framework, so as to accurately extract the group characteristics among individuals within the pedestrian group and the vehicle group. Subsequently, the game interaction behavior between the pedestrian group and the vehicle group is introduced, the group characteristics and the game characteristics between groups are deeply integrated, and finally the deep learning framework is used to capture the behavior intention and the optimal strategy of the pedestrians under the vehicle-pedestrian interaction in the fusion characteristics, so as to accurately judge the behavior intention and generate high-precision pedestrian trajectories.

[0136] Embodiment Two

[0137] This embodiment also discloses a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the steps of the method described in Embodiment One are implemented.

[0138] Embodiment Three

[0139] This embodiment also discloses a computer program product, including a computer program, which when executed by a processor implements the steps of the method described in Embodiment 1.

[0140] The above are only the preferred specific embodiments of the present application, but the protection scope of the present application is not limited thereto. Any changes or substitutions that can be easily thought of by those skilled in the technical field of the present application within the technical scope disclosed by the present application should be covered by the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. A method for pedestrian intention judgment and trajectory generation in vehicle-pedestrian interaction based on group characteristics, characterized in that It includes the following steps: Obtain the vehicle and pedestrian characteristics of the vehicle-pedestrian interaction scenario, calculate the spatio-temporal proximity and collision time through the vehicle and pedestrian characteristics, and identify the pedestrians and vehicles with interaction by preset thresholds of the spatio-temporal proximity and collision time; Specifically, it includes: Obtain the vehicle and pedestrian characteristics of the vehicle-pedestrian interaction scenario; the vehicle and pedestrian characteristics include pedestrian position characteristics, vehicle position characteristics, pedestrian speed characteristics, and vehicle speed characteristics; Calculate the spatio-temporal proximity through the vehicle and pedestrian characteristics and record it in a symmetric spatio-temporal proximity matrix; Calculate the collision time through the vehicle and pedestrian characteristics and record it in a symmetric relative collision time matrix; Based on the spatio-temporal proximity matrix and the relative collision time matrix, set adjustable thresholds respectively, and identify the pedestrians and vehicles with interaction according to the thresholds; Introduce an adjustable distance threshold as a constraint to construct a pedestrian group and a vehicle group that affect pedestrian interaction behavior; construct a deep learning network, and extract the group characteristics of the pedestrian group and the vehicle group respectively; Specifically, it includes: Introduce an adjustable distance threshold as a constraint to respectively screen out the surrounding pedestrians and surrounding vehicles that affect the behavior of interacting pedestrians; Based on the surrounding pedestrians and the surrounding vehicles, introduce interacting pedestrians to construct a pedestrian group and a vehicle group; Respectively construct a scene-adaptive deep learning network, and extract the group characteristics of the pedestrian group and the vehicle group through the deep learning network; According to the group characteristics, simulate the game interaction behavior between the pedestrian group and the vehicle group, and fuse the group characteristics and the game characteristics between the groups to obtain group fusion characteristics; Specifically, it includes: According to the group characteristics, construct a game strategy set for the pedestrian group and the vehicle group; through the game strategy set, establish the game utility of the pedestrian group and the vehicle group; Based on the game utility, simulate the game interaction behavior of the pedestrian group and the vehicle group under vehicle-pedestrian interaction to obtain the expected utility of the pedestrian group and the vehicle group; Based on the expected utility, fuse the group characteristics of the pedestrian group and the vehicle group to obtain group fusion characteristics; According to the group fusion characteristics, perform pedestrian intention judgment and trajectory generation under vehicle-pedestrian interaction through a deep learning framework; Specifically, it includes: Take the group fusion characteristics as the input, construct a behavior intention feature extraction module under the deep learning framework, and extract the behavior intention characteristics in the vehicle-pedestrian interaction scenario; Shape and scale the behavior intention characteristics to obtain the reward function of the interacting pedestrians under vehicle-pedestrian interaction; Take the reward function as the evaluation criterion to obtain the behavior strategy of the interacting pedestrians under vehicle-pedestrian interaction under the influence of the pedestrian group and the vehicle group; Iteratively update the behavior intention feature extraction module and the behavior strategy to perform pedestrian intention judgment in a heterogeneous interaction scenario and generate a pedestrian trajectory.

2. The method according to claim 1, wherein The group characteristics include the internal state relationship of the group and the group position relationship; The calculation formula for the internal state relationship of the group is: ； Among them, and are the state characteristics of each entity within the pedestrian group or vehicle group, and are learnable parameters, is the dimension of the state characteristic; The calculation formula for the group position relationship is: ； Among them, is the spatial distance between the subjects within the pedestrian group or vehicle group; is an adjustable spatial threshold; is an indicator function.

3. The method according to claim 1, wherein The game utility includes the direct risk, indirect risk, avoidance loss, and group utility of collisions; the indirect risk includes the minimum future relative distance, minimum future relative time, and the current speed of the vehicle when a pedestrian or vehicle reaches a potential conflict point.

4. The method according to claim 3, wherein The game utility of the pedestrian group and the vehicle group considering the indirect risk is expressed as: ； wherein, is the utility of the pedestrian group or vehicle group under the strategy combination, is the minimum future relative distance of the pedestrian group or vehicle group under the strategy combination, is the minimum future relative time of the pedestrian group or vehicle group under the strategy combination, is the current speed of the vehicle when the pedestrian group or vehicle group reaches the potential conflict point under the strategy combination.

5. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1-4.

6. A computer program product comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1-4.

Images