Method, apparatus and medium for predicting motion behavior of road participants

Abstract

The embodiment of the present disclosure provides a kind of road participant motion behavior prediction method, device, equipment and medium.The road participant motion behavior prediction method comprises: obtaining the environment perception data collected in M adjacent periods;Adopt motion behavior prediction model to predict the first prediction probability of target road participant executing various motion behaviors, and the second prediction probability of target road participant executing various motion behaviors is predicted using rule-based prediction method;The first prediction probability and the second prediction probability corresponding to various motion behaviors are data fused, and the third prediction probability is obtained;According to the third prediction probability, the predicted motion behavior of target road participant is determined.The scheme provided by the embodiment of the present disclosure fuses the prediction probability of various motion behaviors based on motion behavior prediction model and rule prediction method, so that the prediction of motion behavior has the advantages of both prediction methods, and the prediction method has good adaptability in various scenarios.

Description

TECHNICAL FIELD

[0001] The present disclosure relates to the technical field of automatic driving, in particular to a motion behavior prediction method and device for road participants, equipment and medium. BACKGROUND

[0002] In the automatic driving technology, in order to make reasonable vehicle control decisions, it is necessary to predict the behavior of surrounding road participants in a future period of time, and the prediction result of the behavior of the road participants needs to meet the requirements of real-time and reliability.

[0003] The related technology proposes a method of determining the behavior of road participants based on deep neural network processing of environmental perception data, and a method of determining the behavior of road participants based on rule processing of environmental perception data. However, the two methods have their own defects and cannot meet the above requirements at the same time. Specifically, the method based on deep neural network has poor reliability for short-time prediction results of sudden road participants or sudden behaviors of observed road participants; and the rule processing method cannot fully consider various scenarios, resulting in poor reliability of prediction results for specific scenarios. SUMMARY

[0004] In order to solve the above technical problems, the embodiments of the present disclosure provide a motion behavior prediction method, device, equipment and medium for road participants.

[0005] In a first aspect, the embodiments of the present disclosure provide a motion behavior prediction method for road participants, comprising:

[0006] Obtaining environmental perception data collected in M adjacent periods, M≥2;

[0007] According to at least part of the environmental perception data, a motion behavior prediction model is used to predict a first prediction probability of a target road participant performing various motion behaviors, the motion behavior prediction model is a pre-trained recurrent neural network model, and the motion behavior is a trend behavior of the target road participant relative to a reference object in a target lane;

[0008] According to at least part of the environmental perception data, a rule-based prediction method is used to predict a second prediction probability of the target road participant performing various motion behaviors;

[0009] Data fusion is performed on the first prediction probability and the second prediction probability corresponding to various motion behaviors to obtain a third prediction probability of the target road participant performing various motion behaviors;

[0010] The predicted motion behavior of the target road participant is determined according to the third prediction probability corresponding to various motion behaviors.

[0011] In a second aspect, the embodiments of the present disclosure provide a motion behavior prediction device for a road participant, comprising:

[0012] a data acquisition unit configured to acquire environment perception data collected in M adjacent periods, M≥2;

[0013] a first prediction unit configured to predict, according to at least part of the environment perception data, a first prediction probability of a target road participant performing various motion behaviors by using a motion behavior prediction model, the motion behavior prediction model being a pre-trained recurrent neural network model, and the motion behavior being a trend behavior of the target road participant relative to a reference object in a target lane;

[0014] a second prediction unit configured to predict, according to at least part of the environment perception data, a second prediction probability of the target road participant performing various motion behaviors by using a rule-based prediction method;

[0015] a fusion prediction unit configured to perform data fusion on the first prediction probability and the second prediction probability corresponding to various motion behaviors, to obtain a third prediction probability of the target road participant performing various motion behaviors;

[0016] a motion behavior determination unit configured to determine a predicted motion behavior of the target road participant according to the third prediction probability corresponding to various motion behaviors.

[0017] In a third aspect, the embodiments of the present disclosure provide a computing device, comprising a processor and a memory, the memory being configured to store a computer program; when the computer program is loaded by the processor, the processor is enabled to execute the motion behavior prediction method for a road participant as described above.

[0018] In a fourth aspect, the embodiments of the present disclosure provide a computer readable storage medium, the storage medium storing a computer program, when the computer program is executed by a processor, the processor is enabled to implement the motion behavior prediction method for a road participant as described above.

[0019] The scheme provided by the embodiments of the present disclosure is that, after obtaining the environment perception data, at least part of the environment perception data is processed by using a motion behavior prediction model and a rule-based prediction method respectively, to obtain first prediction probabilities and second prediction probabilities of the target road participant performing various motion behaviors, then the first prediction probabilities and the second prediction probabilities corresponding to the various motion behaviors are fused to obtain third prediction probabilities of the target road participant performing the various motion behaviors, and finally the third prediction probabilities of the various motion behaviors are used to determine the predicted motion behavior of the target road participant. That is, the scheme provided by the embodiments of the present disclosure fuses the prediction probabilities of the various motion behaviors based on the motion behavior prediction model and the rule-based prediction method, so that the prediction of the motion behavior has the advantages of both prediction methods, and the prediction method has good adaptability in various scenarios. BRIEF DESCRIPTION OF DRAWINGS

[0020] The accompanying drawings, which are incorporated herein and form part of the specification, illustrate embodiments consistent with the present disclosure and, together with the description, further serve to explain the principles of the present disclosure.

[0021] In order to more clearly illustrate the technical solutions in the embodiments of the present disclosure or the prior art, the accompanying drawings needed to be used in the embodiments or prior art description will be briefly introduced. Obviously, those skilled in the art can obtain other drawings according to these drawings without any creative effort, in which:

[0022] Figure 1 is a flowchart of a motion behavior prediction method of a road participant provided by the embodiments of the present disclosure;

[0023] Figure 2 is a model architecture diagram of an ensemble learning model provided by an embodiment of the present disclosure;

[0024] Figure 3 is a road scene diagram in some embodiments of the present disclosure;

[0025] Figure 4 is a structure diagram of a motion behavior prediction model used in some embodiments of the present disclosure;

[0026] Figure 5 is a flowchart of a rule-based prediction method used in some embodiments of the present disclosure to determine the second prediction probability;

[0027] Figure 6 is a flowchart of a method for selecting a lateral speed threshold in an embodiment of the present disclosure;

[0028] Figure 7 is a structure diagram of a motion behavior prediction device provided by the embodiments of the present disclosure;

[0029] Figure 8 is a structural schematic diagram of a computing device provided by an embodiment of the present disclosure. DETAILED DESCRIPTION

[0030] Embodiments of the present disclosure will be described in more detail with reference to the drawings. Although some embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure can be implemented in various forms, and should not be interpreted as being limited to the embodiments set forth herein, but rather, these embodiments are provided to more thoroughly and completely understand the present disclosure. It should be understood that the drawings and embodiments of the present disclosure are only for exemplary purposes, and are not intended to limit the scope of protection of the present disclosure.

[0031] The term "comprising" and variations thereof as used herein are open-ended, that is "including but not limited to". The term "based on" is "based, at least in part, on". The term "one embodiment" means "at least one embodiment"; the term "another embodiment" means "at least one additional embodiment"; the term "some embodiments" means "at least some embodiments". Related terms are defined in the following description. It should be noted that the concepts of "first", "second", etc. mentioned in the present disclosure are only used to distinguish different devices, modules or units, and are not intended to limit the order or interdependence of the functions performed by these devices, modules or units.

[0032] It should be noted that the modification of "one" or "multiple" mentioned in the present disclosure is illustrative and not limiting, and those skilled in the art should understand that unless the context clearly indicates otherwise, it should be understood as "one or more".

[0033] To solve the problem that the related art cannot guarantee good reliability in various cases by using a separate processing method to process environmental perception data, an embodiment of the present disclosure provides a new road participant motion behavior prediction method. The new road participant motion behavior prediction method fuses the motion behavior probabilities of road participants predicted by two methods to ultimately determine the predicted motion behavior of the road participants.

[0034] Figure 1 is a flowchart of a road participant motion behavior prediction method provided by an embodiment of the present disclosure. As shown in Figure 1 The road participant motion behavior prediction method provided by an embodiment of the present disclosure includes S110-S150.

[0035] The road participant motion behavior prediction method provided by an embodiment of the present disclosure can be executed by a computing device. In specific implementation, the computing device can be a car machine system of a vehicle, and can also be a remote server or an edge server. In the following, the car machine system of the vehicle is taken as the computing device in the embodiments of the present disclosure to analyze the motion behavior prediction method provided by the embodiments of the present disclosure.

[0036] S110: Obtain environment perception data collected in M adjacent periods, M≥2.

[0037] The environment perception data is obtained by using an environment perception sensor to perceive the road environment, and is data representing the motion characteristics of road participants or data representing the characteristics of reference objects in the road. Correspondingly, the environment perception data collected in the adjacent periods is data representing the motion characteristics of road participants or the characteristics of reference objects in the road obtained by using an environment perception sensor to perceive the road environment in the adjacent periods.

[0038] The aforementioned reference object can be a position-fixed road marker such as a lane line, a lane center line, etc.

[0039] In a specific implementation, the motion characteristic data can include speed, acceleration, and position. The speed can include longitudinal speed (speed along the direction of road extension) and lateral speed (speed perpendicular to the direction of road extension), the acceleration can include longitudinal acceleration (acceleration along the direction of road extension) and lateral acceleration (acceleration perpendicular to the direction of road extension), and the position can include longitudinal position (position along the direction of road extension) and lateral position (position perpendicular to the direction of road extension).

[0040] Since the motion behavior prediction model is used in the subsequent method steps to predict the prediction probability (hereinafter referred to as the first prediction probability) of the target road participant performing various motion behaviors based on the environment perception data collected in the adjacent periods, and the motion behavior prediction model is a recurrent neural network model for processing time series data, in order to ensure that the motion behavior prediction model can realize the prediction of the probability of various motion behaviors, it is necessary to set the number of adjacent periods M to at least 2.

[0041] In a specific implementation, in order to enable the motion behavior prediction model to accurately determine the first prediction probability of the target road participant performing various motion behaviors in the subsequent steps, M should be set reasonably. In one specific embodiment, when the sampling period is 0.25 s, M can be set to 10.

[0042] The aforementioned environment perception sensor is a sensor deployed on the ego vehicle or on the roadside, and is preferably a sensor deployed on the ego vehicle. The environment perception sensor can include a camera, a lidar, an inertial measurement unit, a millimeter wave radar, and an ultrasonic radar, and the embodiments of the present disclosure are not particularly limited. In view of the fact that the motion behavior prediction method for road participants in the embodiments of the present disclosure is mostly used to predict the motion behavior of a target road participant at a relatively long distance, in order to ensure the accuracy of the behavior prediction of the target road participant from the perspective of the data source, the environment perception sensor is preferably a camera, a lidar, and an inertial measurement unit, etc. in actual implementation.

[0043] S120: According to at least part of the environment perception data, a motion behavior prediction model is used to predict a first prediction probability of the target road participant performing various motion behaviors.

[0044] In the embodiments of the present disclosure, the motion behavior prediction model is a neural network model trained using sample environment perception data and used to predict the probability of the target road participant performing various motion behaviors. The motion behavior prediction model is a recurrent neural network model, that is, a recurrent neural network model that can recursively chain process environment perception data sorted in chronological order.

[0045] The target road participant is a road participant currently being focused on and whose future period motion behavior needs to be predicted. The target road participant can be a non-vulnerable road participant on the road (such as a conventional four-wheeled vehicle), or a vulnerable road participant on the road (such as a pedestrian, a bicycle, or a motorcycle). In view of the fact that the method provided in the embodiments of the present disclosure is mostly applied to a main road on which a non-vulnerable road participant travels, the target road participant is mostly a non-vulnerable road participant.

[0046] The motion behavior is the tendency behavior of the target road participant relative to the target lane in which the reference object is located. The motion behavior can include parallel straight driving, lane cut-in, and lane cut-out.

[0047] Parallel straight driving refers to a motion behavior of moving generally along the extension direction of the target lane. At this time, the target road participant can be located inside the target lane or outside the target lane, and the specific location needs to be determined according to the positional relationship between the preceding target road participant and the target lane.

[0048] Lane cut-in refers to a motion behavior of the target road participant cutting into the target lane from an adjacent lane at a certain cut-in angle. After the execution of the lane cut-in is completed, the target road participant will move to the target lane. It should be noted that the target road participant can not have moved to the target lane during part of the process of performing the lane cut-in motion behavior.

[0049] Lane cut-out refers to the movement behavior of the target road participant moving out of the target lane at a certain cut-out angle. After the execution of the lane cut-out, the target road participant will move out of the target lane. It should be noted that although the target road participant performs the movement behavior of lane cut-out, it may not have moved out of the target lane during the execution of the movement behavior.

[0050] According to the foregoing analysis, the first prediction probability is a probability of the target road participant performing parallel straight driving, lane cut-in and lane cut-out, respectively.

[0051] It should be noted that in S120, the computing device predicts the first prediction probability of the target road participant performing various movement behaviors according to at least part of the environment perception data by using the movement behavior prediction model. At least part of the foregoing environment perception data includes at least part of the movement characteristics of the target road participant, and may also include part of the movement characteristics of the related road participant, which will be analyzed later. Here, it is only necessary to understand and confirm that after the movement behavior prediction model processes at least part of the foregoing environment perception data, the first prediction probability of the target road participant performing various movement behaviors can be obtained.

[0052] S130: Predict the second prediction probability of the target road participant performing various movement behaviors according to at least part of the environment perception data by using the rule-based prediction method.

[0053] The rule-based prediction method predicts the prediction probability of the target road participant performing various movement behaviors by processing at least part of the environment perception data according to the predetermined decision rule to determine the prediction probability. The foregoing decision rule can be a movement behavior probability prediction rule determined according to experience. How to predict the second prediction probability of the target road participant performing various movement behaviors by using the rule-based prediction method will be analyzed later.

[0054] It should be noted that the "at least part of the environment perception data" processed by the rule-based prediction method may be the same as or different from the "at least part of the environment perception data" processed by the movement behavior prediction model in the foregoing. It needs to be determined according to the actual situation. In most cases, they are not the same, which will be analyzed later.

[0055] S140: Data fusion is performed on the first prediction probability and the second prediction probability corresponding to various movement behaviors to obtain the third prediction probability of the target road participant performing various movement behaviors.

[0056] After obtaining the first prediction probability and the second prediction probability corresponding to each motion behavior, the first prediction probability and the second prediction probability of each motion behavior can be fused to obtain a third prediction probability of each motion behavior.

[0057] In a specific implementation, the method of data fusion of the first prediction probability and the second prediction probability can be various.

[0058] In one specific application, the computing device can perform weighted average on the first prediction probability and the second prediction probability corresponding to the motion behavior, and take the obtained average value as the third prediction probability. For example, in the case that the first prediction probability and the second prediction probability of parallel straight driving are 0.41 and 0.47 respectively, the third prediction probability obtained by performing weighted average on the first prediction probability and the second prediction probability is 0.44 (at this time, the weights of the first prediction probability and the second prediction probability are both 0.5, of course, the weights of the first prediction probability and the second prediction probability can also be adaptively determined as other values according to actual conditions).

[0059] In another specific embodiment, the computing device can use an ensemble learning model to perform data fusion processing on the first prediction probability and the second prediction probability corresponding to each motion behavior to obtain the third prediction probability of the target road participant performing each motion behavior.

[0060] The ensemble learning model is a trained neural network model. After obtaining the first prediction probability and the second prediction probability of each motion behavior, the first prediction probability and the second prediction probability corresponding to each motion behavior are spliced into a probability vector in a pre-set order, and then the probability vector is input into the ensemble learning model to obtain the third prediction probability of the target road participant performing each motion behavior.

[0061] Figure 2 is a model architecture diagram of the ensemble learning model provided by one embodiment of the present disclosure. As Figure 2 shown, the ensemble learning model 200 provided by the embodiment of the present disclosure includes a full connection layer one 201, a full connection layer two 202, and a normalization function layer 203. After the probability vector is input into the input port of the full connection layer one 201, after the processing of the full connection layer one 201, the full connection layer two 202, and the normalization function layer 203, the third prediction probability of each motion behavior can be obtained.

[0062] S150: determining the predicted motion behavior of the target road participant according to the third prediction probability corresponding to each motion behavior.

[0063] After obtaining the third motion probabilities of various motion behaviors, the computing device can determine the maximum probability among the third prediction probabilities corresponding to various motion behaviors, and take the maximum probability as the predicted motion behavior of the target road participant, that is, take the motion behavior corresponding to the maximum third prediction probability as the predicted motion behavior of the target road participant.

[0064] The motion behavior prediction method for a road participant provided in the embodiments of the present disclosure can be used to process at least part of the environment perception data by using the motion behavior prediction model and the rule-based prediction method respectively after obtaining the environment perception data, to obtain the first prediction probability and the second prediction probability of various motion behaviors performed by the target road participant, and then perform data fusion on the first prediction probability and the second prediction probability corresponding to various motion behaviors to obtain the third prediction probability of various motion behaviors performed by the target road participant, and finally determine the predicted motion behavior of the target road participant based on the third prediction probability of various motion behaviors.

[0065] That is, the motion behavior prediction method for a road participant provided in the embodiments of the present disclosure fuses the prediction probabilities of various motion behaviors based on the motion behavior prediction model and the rule-based prediction method, so that the prediction of the motion behavior has both the adaptability to the scene based on the motion behavior prediction model and the advantages of high accuracy and good interpretability of the output result based on the rule-based method, and thus the prediction method has good adaptability in various scenes.

[0066] As analyzed above, in S120, at least part of the environment perception data is processed by using the motion behavior prediction model to obtain the first prediction probability of various motion behaviors performed by the target road participant.

[0067] In some embodiments of the present disclosure, the environment perception data includes the first motion feature data of the target road participant, the second motion feature data of the related road participant, and the position of the reference object. The aforementioned related road participant is a road participant directly affecting the motion behavior of the target road participant.

[0068] Figure 3 FIG. 1 is a schematic diagram of a road scene in some embodiments of the present disclosure. As shown in FIG. 1, the road scene includes a target road participant 101, a related road participant 102, and a reference object 103. Figure 3As shown, the road scene is a three-lane scene in the same direction, in which the target road participant drives in the middle lane. In this scenario, the relevant road participants include vehicle a in front of the target road participant, vehicle b in the front left of the target road participant, vehicle c in the back left of the target road participant, vehicle d in the front right of the target road participant, and vehicle e in the back right of the target road participant. It should be noted that according to the road traffic regulations, the target road participant has a more optimal right of way with respect to the vehicle f behind it, so in a normal case, the vehicle f is not a road participant directly affecting the motion behavior of the target road participant.

[0069] In the embodiments of the present disclosure, the reference is the position of the lane line (including the left lane line and the right lane line) at the position of the target road participant.

[0070] Of course, in actual cases, the road scene may not be the aforementioned three-lane scene in the same direction, but a two-lane scene in the same direction or a single-lane scene, and the relevant road participants may only be part of the aforementioned vehicles a-e, at which time the number of relevant road participants will also be adaptively reduced.

[0071] In addition, in actual scenarios, there may be only one side lane line at the position of the target road participant (for example, the target road participant drives in a two-lane in the same direction, which is separated by a middle lane line), at which time the reference is only the position of the aforementioned one side lane line.

[0072] In the road scene as shown in Figure 3 The S121 may specifically include the following S121.

[0073] S121: According to at least part of the first motion feature data in the M adjacent periods, at least part of the second motion feature data in the M adjacent periods, and the position of the reference, the motion behavior prediction model is used to predict the first prediction probability of the target road participant performing various motion behaviors.

[0074] That is, in the embodiments of the present disclosure, the motion behavior prediction model mines the factors affecting the target road participant's decision of motion behavior according to at least part of the first motion feature data of the target road participant in the M adjacent periods, at least part of the second motion feature data of the relevant road participants affecting the motion behavior of the target road participant in the M adjacent periods, and the position of the reference, and predicts the first prediction probability of the target road participant performing various motion behaviors based on the aforementioned factors.

[0075] In some specific applications, the first motion characteristic data of the target road participant includes a first longitudinal speed, a first longitudinal acceleration, a first lateral speed and a first lateral position, the second motion characteristic data of the related road participant includes a second longitudinal speed and a second longitudinal acceleration, and the position of the reference object includes a lateral position of the reference object.

[0076] In the embodiments of the present disclosure, the reference object is a lane line of a lane in which the target road participant is located, and the lateral position of the corresponding reference object is the lateral position of the lane line on both sides of the position of the target road participant. The lateral position of the lane line is analyzed as follows.

[0077] In some embodiments of the present disclosure, the sensor for environment perception includes a camera. The road image formed by the camera when shooting the road includes lane line information. After obtaining the road image, the computing device determines the lane line parameter equation in a manner such as inverse perspective processing on the road image. In one specific embodiment, the lane line parameter equation is a cubic polynomial, and the expression is y=a0+a1x+a2x 2 +a3x 3 , wherein a0, a1, a2 and a3 are lane line parameters, and x is a longitudinal distance to be input. After obtaining the lane line parameter equation and obtaining the longitudinal distance x between the target road participant and the camera, the x is brought into the lane line parameter equation, and the lateral position y of the reference object can be obtained. It should be noted that the lateral position of the lane line in the embodiments of the present disclosure is the lateral position of the lane line related to the position of the target road participant in each adjacent period.

[0078] In the case of determining the lateral position of the lane line, the foregoing S121 includes S1211-S1212.

[0079] S1211: calculating a first lateral distance in each adjacent period according to the first lateral position and the lateral position of the reference object in each adjacent period.

[0080] In the embodiments of the present disclosure, the first lateral distance is the distance between the target road participant and the target lane center line. The relative position of the target road participant and the target lane can be represented according to the first lateral distance.

[0081] In a specific implementation, the first lateral distance in each adjacent period is calculated according to the first lateral position and the lateral position of the reference object in each adjacent period as follows: assuming that the lane line (that is, the reference object) includes a left lane line and a right lane line, the lateral position of the left lane line is y left , the lateral position of the right lane line is y right , and the first lateral position of the target road participant is y targetDist_2_Left_Lane = y target -y left Dist_2_Right_Lane = y target -y target It should be noted that in order to reflect the left and right relationship of the target road participant relative to the lane line, the aforementioned Dist_2_Left_Lane and Dist_2_Right_Lane can be positive or negative, which needs to be determined according to the coordinate system setting and the relative relationship between the target road participant and the target lane. After determining the aforementioned Dist_2_Left_Lane and Dist_2_Right_Lane, the first lateral distance DTCL = (Dist_2_Left_Lane + Dist_2_Right_Lane) / 2 can be obtained.

[0082] S1212: The first longitudinal velocity, the second longitudinal velocity, the first longitudinal acceleration, the second longitudinal acceleration, the first lateral velocity and the corresponding first lateral distance in the M adjacent periods are processed by using the motion behavior prediction model to obtain the first prediction probability.

[0083] After obtaining the first lateral distance of the M adjacent periods, the first lateral distance of the aforementioned M periods is then respectively combined with the first longitudinal velocity, the second longitudinal velocity, the first longitudinal acceleration, the second longitudinal acceleration and the first lateral velocity of the corresponding period to be sequentially input into the motion behavior prediction model, that is, the first prediction probability corresponding to each motion behavior can be obtained.

[0084] The following analyzes how to use the motion prediction behavior model to process the motion feature vector to obtain the first prediction probability.

[0085] Figure 4 is a structural schematic diagram of the running behavior prediction model 400 adopted by some embodiments of the present disclosure. As shown in Figure 4 The running behavior prediction model 400 in the embodiments of the present disclosure includes a first input full connection network 401, a second input full connection network 402, a recurrent neural network 403 and an output full connection network 404.

[0086] The first input fully connected network 401 is configured to extract and abstract the first longitudinal velocity, the second longitudinal velocity, the first longitudinal acceleration and the second longitudinal acceleration of each adjacent period to obtain a first intermediate vector. The first intermediate vector is configured to represent the implicit features of the relative relationship between the target road participant and the relevant road participant in the longitudinal direction. In some embodiments, the first input fully connected network 401 includes five sub-networks, which are parallel processing networks, and each of the five sub-networks is configured to process the first longitudinal velocity, the first longitudinal acceleration of the target road participant and the second longitudinal velocity and the second longitudinal acceleration of one of the aforementioned relevant road participants a-e to obtain the first intermediate vector. Each of the five sub-networks includes a first fully connected layer (L1-1, L1-2, L1-3, L1-4 and L1-5, respectively) and a second fully connected layer (L2-1, L2-2, L2-3, L2-4 and L2-5, respectively). The network dimension of the first fully connected layer is higher than the dimension of the corresponding input vector (i.e., the dimension of the input data), which serves to increase the dimension of the features to extract as many implicit features as possible from the corresponding input data. The network dimension of the second fully connected layer is lower than that of the first fully connected layer, which serves to abstract the features of the increased dimension.

[0087] The second input fully connected network 402 is configured to extract and abstract the first lateral velocity and the corresponding first lateral distance of each adjacent period to obtain a second intermediate vector corresponding to each adjacent period. The second intermediate vector determines the features of the movement of the target road participant in the lateral direction. In one embodiment, the second input fully connected network 402 includes a fully connected layer (L1-6).

[0088] After obtaining the first intermediate vector and the second intermediate vector corresponding to each adjacent period, the computing device can further concatenate the first intermediate vector and the second intermediate vector to obtain a concatenated vector.

[0089] The recurrent neural network 403 is configured to sequentially process the concatenated vectors corresponding to each adjacent period in chronological order to obtain a corresponding intermediate vector, and feed the aforementioned intermediate vector to the input port of itself to continue processing the concatenated vector corresponding to the next period until the third intermediate vector corresponding to the last period is obtained.

[0090] The output fully connected network 404 is configured to process the third intermediate vector to obtain an output vector, and the elements in the output vector implicitly represent the probabilities of the target road participant performing various motion behaviors. After obtaining the output vector, the first prediction probabilities of various motion behaviors can be obtained after the output vector is processed by a normalization exponential function.

[0091] According to the above explanation of the structure of the motion behavior prediction model, the S1212 can include the following A-E.

[0092] A: The first input full connection network is used to process the first longitudinal velocity, the second longitudinal velocity, the first longitudinal acceleration and the second longitudinal acceleration corresponding to each adjacent period, to obtain a first intermediate vector corresponding to each adjacent period.

[0093] B: The second input full connection network is used to process the first lateral velocity and the corresponding first lateral distance in each adjacent period, to obtain a second intermediate vector corresponding to each adjacent period.

[0094] C: The first intermediate vector and the second intermediate vector corresponding to each adjacent period are spliced to obtain a spliced vector.

[0095] D: According to the period order of the adjacent periods, the spliced vector corresponding to each adjacent period is processed in sequence by using the recurrent neural network to obtain a third intermediate vector.

[0096] E: The output full connection network is used to process the third intermediate vector to obtain an output vector, and the first prediction probability of the target road participant performing various motion behaviors is determined based on the output vector.

[0097] In the embodiments of the present disclosure, the recurrent neural network can be a network based on long short-term memory (Long Short-Term Memory, LSTM). The long short-term memory network gives different weights to the data processed by different adjacent periods through the internally set forgetting gate, input gate and output gate, and takes into account the influence of the input data of the near period and the far period on the first prediction probability, and makes the influence of the input data of the near period on the first prediction probability greater, and the influence of the input data of the far period on the first prediction probability smaller, so as to make the accuracy of the obtained first prediction probability higher.

[0098] Of course, the recurrent neural network is not limited to the aforementioned network based on long short-term memory, and can also be other types of networks with sequence data processing function.

[0099] In the foregoing analysis, it is mentioned that the first intermediate vector is obtained by processing the first longitudinal velocity, the second longitudinal velocity, the first longitudinal acceleration and the second longitudinal acceleration corresponding to each adjacent period by using the first fully connected network, and the second intermediate vector is obtained by processing the first lateral velocity and the corresponding first lateral distance corresponding to each adjacent period by using the second fully connected network. However, the foregoing input data has different measurement units, and the orders of magnitude of the data are different, which causes inconvenience in the training and use of the motion behavior prediction model, for example, causes unreasonable setting of some node parameters, and the network model is complex. In addition, if the numerical values of the sample data are random, the calculation amount and convergence during model training may be poor. In order to avoid the foregoing problems, in the embodiments of the present disclosure, before the motion behavior prediction model is trained by using the sample data and before the model is used, the input data is pre-processed, encoded into specific data elements, and the foregoing data elements are input as real data to the motion behavior prediction model.

[0100] In one specific embodiment, for the first longitudinal velocity v obj and the second longitudinal velocity v, if v obj -v>3, the first longitudinal velocity v obj and the second longitudinal velocity v are jointly encoded as [1, 0]; if v obj -v<-3, the first longitudinal velocity v obj and the second longitudinal velocity v are jointly encoded as [0, 1]; if -3≤v obj -v<3, the first longitudinal velocity v ob j and the second longitudinal velocity v are jointly encoded as [1, 1]; if v obj and the second longitudinal velocity v are jointly encoded as [0, 0].

[0101] In one specific embodiment, for the first longitudinal acceleration a obj and the second longitudinal acceleration a, if a obj -a>0.2, the first longitudinal acceleration a obj and the second longitudinal acceleration a are jointly encoded as [1, 0]; if a obj -a<-0.2, the first longitudinal acceleration a obj and the second longitudinal acceleration a are jointly encoded as [1, 0]; if -0.2≤a obj -a≤0.2, the first longitudinal acceleration a obj and the second longitudinal acceleration a are jointly encoded as [1, 1]; if a obj and the second longitudinal acceleration a are jointly encoded as [0, 0].

[0102] In one embodiment, for the first lateral velocity v y If |v y |≥3, the corresponding encoded data is v y / |v y |, if |v y |<3, the corresponding encoded data is v y / 3.

[0103] In one embodiment, for the first lateral distance DTCL, if |DTCL|≥8, the corresponding encoded data is DTCL / |DTCL|, if |DTCL|<8, the corresponding encoded data is DTCL / 8.

[0104] As mentioned above, in some embodiments of the present disclosure, the first input full connection network includes five sub-networks, and each of the five sub-networks is used to process data of a corresponding related road participant. However, in actual road scenarios, there can be no corresponding related road participant at the corresponding position, and in this case, the data of the corresponding related road participant can be directly set to 0 to realize the operation and use of the model.

[0105] As mentioned above, the use of the motion behavior prediction model is mentioned in the foregoing embodiments, and how to train the motion behavior prediction model will be analyzed below.

[0106] As described above, the environment perception data can include the first motion feature data of the target road participant and the lateral position of the reference object. In this case, in some embodiments, the foregoing S130 can include the following S131, which predicts the second prediction probability of the target road participant performing various motion behaviors according to at least part of the environment perception data using a rule-based prediction method.

[0107] S131: obtaining the second prediction probability of the target road participant performing various motion behaviors in the current period according to the first motion feature data and the lateral position of the reference object in at least part of the M adjacent periods using a rule-based prediction method.

[0108] Figure 5 is a flowchart of determining the second prediction probability using a rule-based prediction method in some embodiments of the present disclosure. As shown in Figure 5 , in some embodiments, the computing device can use the following S310-S360 when performing the foregoing S131.

[0109] S310: determining the relative position of the target road participant according to the first lateral position of the nearest adjacent period and the lateral position of the reference object.

[0110] The relative position is the position of the target road participant relative to the target lane.

[0111] As analyzed above, the reference object can be a lane line of the target lane. According to the first lateral position corresponding to the last approaching period and the lateral position of the reference object, the relative position of the target lane is determined, i.e., the position of the target road participant relative to the target lane is determined, specifically, whether the target road participant is located in the target lane or in the adjacent lane outside the target lane. Of course, if there is no lane line in the road, the reference object can also be the ego vehicle, i.e., the relative position of the target road participant is determined by the lateral positions of the ego vehicle and the target road participant.

[0112] S320: Obtain the target confidence and the corresponding target lateral speed threshold in the lookup table.

[0113] The lookup table is determined according to empirical data and is used to determine the probabilities of various motion behaviors of the target road participant. The lookup table can determine lateral speed thresholds of different sizes and their corresponding target confidences.

[0114] The aforementioned target confidence is used to represent the confidence of the target road participant performing a specific motion behavior within the corresponding target lateral speed threshold range, i.e., the probability of the target road participant performing the specific motion behavior.

[0115] Specifically, how to obtain the target confidence and the corresponding target lateral speed threshold in the lookup table needs to be determined according to the type of the header information of the lookup table, which will be expanded later. Here, it is only necessary to determine that the target lateral speed threshold and its corresponding target confidence can be obtained from the lookup table.

[0116] S330: Compare the size of the first lateral speed corresponding to the last approaching period and the target lateral speed threshold to obtain a first comparison result.

[0117] The last approaching period is the last period in the M approaching periods. The first comparison result obtained by comparing the size of the first lateral speed corresponding to the last approaching period and the target lateral speed threshold is used to represent whether the first lateral speed is greater than the target lateral speed threshold or less than the target lateral speed threshold.

[0118] S340: Determine the first motion behavior corresponding to the target confidence according to the first comparison result, the direction of the first lateral speed, and the relative position.

[0119] After obtaining the first comparison result, i.e., determining whether the first lateral speed is greater than the target lateral speed threshold or less than the target lateral speed threshold, the first motion behavior corresponding to the target confidence can be determined according to the first comparison result, the direction of the first lateral speed, and the relative position.

[0120] For example, if the first comparison result is that the first lateral velocity is greater than the target lateral velocity threshold, the direction of the first lateral velocity is to the right, and the relative position indicates that the target road participant is located within the target lane, it is determined that the target road participant has a tendency to move to the right and exit the target lane, and the corresponding first motion behavior is lane-changing cut-out.

[0121] For another example, if the first comparison result is that the first lateral velocity is less than the target lateral velocity threshold, the direction of the first lateral velocity is to the right, and the relative position indicates that the target road participant is located within the target lane, it is determined that the target road participant has a tendency to continue driving along the target lane, and the corresponding first motion behavior is parallel straight driving.

[0122] For another example, if the first comparison result is that the first lateral velocity is greater than the target lateral velocity threshold, the direction of the first lateral velocity is to the right, and the relative position indicates that the target road participant is located to the left of the target lane, it is determined that the target road participant has a tendency to drive to the right into the target lane, and the corresponding first motion behavior is lane-changing cut-in.

[0123] For another example, if the first comparison result is that the first lateral velocity is less than the target lateral velocity threshold, the direction of the first lateral velocity is to the right, and the relative position indicates that the target road participant is located to the left of the target lane, it is determined that the target road participant has a tendency to drive to the right into the target lane, and the corresponding first motion behavior is parallel execution.

[0124] S350: Taking the target confidence as a second prediction probability corresponding to the first motion behavior.

[0125] After determining the type of the first motion behavior, the subsequent computing device can directly take the aforementioned target confidence as a confidence corresponding to the first motion behavior, that is, as a second prediction probability corresponding to the first motion behavior.

[0126] S360: According to the second prediction probability corresponding to the first motion behavior, determining a second prediction probability corresponding to other motion behaviors.

[0127] According to the knowledge of probability theory, the sum of the probabilities of the target road participant performing various motion behaviors is 1. After determining the second prediction probability of the target road participant performing the first motion behavior, the sum of the second prediction probabilities of the target road participant performing other motion behaviors can also be determined. Accordingly, according to the sum of the second prediction probabilities of the target road participant performing other motion behaviors, the probability of the target road participant performing other motion behaviors can be estimated. Accordingly, S360 can include S361-S362.

[0128] S361: Determining a residual probability according to the second prediction probability corresponding to the first motion behavior.

[0129] S362: determining a second prediction probability of the other motion behavior according to the number of the other motion behaviors except the first motion behavior and the residual probability.

[0130] In the case that the second prediction probability of the first motion behavior is ES_Conf, the residual probability is 1-ES_Conf.

[0131] In some embodiments of the present disclosure, after determining the residual probability, the residual probability can be divided by the number of the other motion behaviors to obtain the second prediction probability of the other motion behavior. According to the foregoing analysis, the number of the other motion behaviors is 2, and the second prediction probability of the corresponding other motion behavior is (1-ES_Conf) / 2.

[0132] In some other embodiments of the present disclosure, in a relatively short time, the motion behavior of the target road participant can only switch between two behaviors, and therefore, after determining the residual probability, the residual probability can be taken as the second prediction probability of one of the other motion behaviors, and the second prediction probability of the other of the other motion behaviors is set to 0. For example, if it is determined that the first motion behavior is lane-changing cut-in, the residual probability is determined as the second prediction probability of parallel straight driving, and the second prediction probability of lane-changing cut-out is set to 0.

[0133] In some embodiments of the present disclosure, one target confidence in the lookup table can correspond to multiple lateral speed thresholds. In this case, when performing S320, a suitable target lateral speed threshold needs to be selected from the multiple lateral speed thresholds for performing the comparison step of subsequent S330.

[0134] Figure 6 is a method flowchart for selecting a target lateral speed threshold in embodiments of the present disclosure. As shown in Figure 6 In order to be able to select a suitable target lateral speed threshold from the multiple lateral speed thresholds, in some embodiments, before performing the foregoing S320, the computing device can further perform S4100-S450.

[0135] S410: comparing whether the time length of the M adjacent periods is greater than a first preset time length; if not, performing S420; if yes, performing S430.

[0136] S420: taking a first set threshold in the multiple lateral speed thresholds as the target lateral speed threshold.

[0137] The first preset time length is a time length preset for determining whether the target road participant is a long-time existing target or a short-time existing target. If the time length of the M adjacent periods (i.e., the time length in which the target road participant exists) is greater than the first preset time length, it indicates that the target road participant is a long-time existing target. Conversely, if the time length of the M adjacent periods is less than or equal to the first preset time length, it indicates that the target road participant is a short-time existing target.

[0138] In the embodiments of the present disclosure, in the case where the target road participant is a short-time existing target, the historical information for predicting the motion behavior of the target road participant is not much, and at this time, the motion behavior of the target road participant can only be predicted based on short-time data.

[0139] In the embodiments of the present disclosure, if the target road participant is a short-time motion target, a first preset threshold value in the plurality of lateral velocity threshold values is taken as a target lateral velocity threshold value. The first preset threshold value is a threshold value that is relatively large compared with other preset threshold values. By setting the first preset threshold value to be relatively large, that is, by setting the comparison value for determining the first lateral velocity to be relatively large, the target road participant can be more reliably determined to have a high probability of performing a specific motion behavior.

[0140] S430: According to the first lateral position of the target road participant and the lateral position of the reference object in the N adjacent periods, the corresponding lateral distances are calculated.

[0141] In the case where the target road participant is a long-time existing target, the historical information for predicting the motion behavior of the target road participant is relatively rich, and at this time, the motion behavior of the target road participant can be more accurately predicted by using more data.

[0142] In the embodiments of the present disclosure, in the case where it is determined that the target road participant is a long-time motion target, the computing device can obtain the first lateral position of the target road participant and the lateral position of the reference object in the N adjacent periods of the M adjacent periods, and calculate the corresponding lateral distances using the first lateral position and the lateral position of the reference object in the corresponding adjacent period. It should be noted that M≥N≥3.

[0143] S440: According to the lateral distances of adjacent periods in the N adjacent periods, N-1 lateral distance change values are obtained, and the positive and negative signs of the N-1 lateral distance change values are determined.

[0144] In the embodiments of the present disclosure, M≥N≥3.

[0145] After the lateral distances corresponding to the N adjacent periods are determined, the lateral distances of adjacent periods are subtracted (specifically, the lateral distance of the latter period is subtracted from the lateral distance of the former period), which can obtain N-1 lateral distance change values. The aforementioned lateral distance change value can be positive or negative. If the distance change value is positive, it indicates that the target road participant is moving away from the target lane; if the distance change value is negative, it indicates that the target road participant is moving close to the target lane.

[0146] S450: In the case where the signs of the N-1 distance change values are the same, a second set threshold value in the plurality of lateral speed threshold values is taken as the target lateral speed threshold value; in the case where the signs of the N-P-1 distance change values corresponding to the last N-P adjacent periods are the same but the sign of the N-1 distance change value is not the same, a third set threshold value is taken as the target lateral speed threshold value.

[0147] It should be noted that P≤N-3. In the embodiment of the present disclosure, in the case of N-1 distance change values, the N-1 distance change values can be used to determine the motion behavior of the target road participant.

[0148] If the signs of the N-1 distance change values are the same, it indicates that the target road participant has a tendency to continuously move in a certain direction according to a certain motion behavior, and accordingly, even if the first lateral speed of the target road participant is small, the probability of its executing the aforementioned specific motion behavior can be determined more accurately, and at this time, the second set threshold value can be taken as the threshold value for comparison with the first lateral speed. It should be noted that the second set threshold value is smaller than the first set threshold value, so that the second prediction probability of the target road participant executing the specific motion behavior can be determined more reliably in the case where the first lateral speed is small.

[0149] Similarly, if the N-P-1 distance change values corresponding to the last N-P adjacent periods are the same but the sign of the N-1 distance change value is not the same, it can also be considered with a certain degree of confidence that the target road participant has a tendency to continuously move in a certain direction according to a certain motion behavior. At this time, the third set threshold value can be taken as the threshold value for comparison with the first lateral speed. However, since the N-P-1 distance change values corresponding to the aforementioned N-P adjacent periods are the same but the sign of the N-1 distance change value is not the same, the reliability is not as high as that in the case where the signs of the N-1 distance change values are the same, and accordingly, in order to adapt to the decrease in the aforementioned reliability, the third set threshold value is greater than the second set threshold value, but the third set threshold value is still smaller than the first set threshold value.

[0150] In some embodiments, the distance of the target road participant relative to the ego vehicle (i.e., the target vehicle) also affects the determination and selection of the target confidence and the corresponding target lateral speed threshold in the lookup table. In this case, the aforementioned S320 of obtaining the target confidence and the corresponding target lateral speed threshold in the lookup table can be S321 if the first motion feature data further includes a first longitudinal distance of the target road participant relative to the target vehicle.

[0151] S321 : retrieving the lookup table according to the first longitudinal distance to obtain the target confidence and the target lateral speed threshold associated with the first longitudinal distance.

[0152] Table 1 is a lookup table of target confidence and target lateral speed threshold. As shown in Table 1, the corresponding target confidence and target lateral speed threshold are different for the target road participant at a far distance, a middle distance, and a near distance.

[0153] Table 1: Lookup table of target confidence and target lateral speed threshold

[0154]

[0155] In some embodiments, the road segment where the target road participant is currently located also affects the determination and selection of the target confidence and the corresponding target lateral speed threshold in the lookup table. In this case, before performing the aforementioned S320 of obtaining the target confidence and the corresponding target lateral speed threshold in the lookup table, the computing device can further perform the following S370.

[0156] S370: obtaining a road type of a road where the target road participant travels,

[0157] In a specific implementation, the road type is an intersection or a road segment. The road segment type of the road where the target road participant travels can be determined by processing the sensor-detected environmental perception signals, such as by processing the road images captured by a camera.

[0158] In the case of performing the aforementioned S370, the aforementioned S320 of obtaining the target confidence and the corresponding target lateral speed threshold in the lookup table can be S322.

[0159] S322: retrieving the lookup table according to the road type to obtain the target confidence and the target lateral speed threshold associated with the road type.

[0160] Table 2 is a lookup table of target confidence and target lateral speed threshold. As shown in Table 2, the corresponding target confidence and target lateral speed threshold are different for the target road participant at a far distance, a middle distance, and a near distance when the road type where the target road participant is located is an intersection or a road segment.

[0161] Table 2 Target Confidence and Target Lateral Velocity Threshold Lookup Table

[0162]

[0163]

[0164] In some embodiments, in case the road is a non-closed road (e.g. not a highway or a city main road), the type of the target road participant can not only be a vehicle, but also a vulnerable road participant such as a pedestrian, a bicycle or a motorcycle. Different types of target road participants correspond to different target confidence and target lateral velocity threshold. In this case, before performing the aforementioned S320 to obtain the target confidence and the corresponding target lateral velocity threshold from the lookup table, the computing device can further perform the following S370.

[0165] S370: Determine the participant type of the target road participant, which is either a vehicle or a non-vehicle.

[0166] In case the aforementioned S370 is performed, the aforementioned S320 to obtain the target confidence and the corresponding target lateral velocity threshold from the lookup table can be S323.

[0167] S323: Obtain the target confidence and the target lateral velocity threshold associated with the participant type from the lookup table according to the participant type.

[0168] Table 3 is a target confidence and target lateral velocity threshold lookup table. As shown in Table 3, for different types of target road participants, the corresponding target confidence and target lateral velocity threshold are not the same.

[0169] Table 3 Target Confidence and Target Lateral Velocity Threshold Lookup Table

[0170]

[0171] It should be noted here that it is not contradictory to determine the target confidence and the target lateral velocity threshold by the first longitudinal distance, the road type of the road on which the target road participant travels and the type of the target road participant. In actual applications (e.g. Table 3), the target confidence and the target lateral velocity threshold can be determined by the three parameters at the same time.

[0172] In addition, it should also be noted that the aforementioned S321, S322 and S323 are not contradictory to S410-S460. S321, S32 or S323 can be performed to determine the target confidence and the target lateral velocity threshold at the same time when the aforementioned S410-S460 selection method is performed.

[0173] The foregoing describes the method for predicting the motion behavior of the road participant provided by the embodiments of the present disclosure. The following describes how to train the motion behavior prediction model and the ensemble learning model used in the embodiments of the present disclosure.

[0174] In the embodiments of the present disclosure, since the motion behavior prediction model and the ensemble learning model are both neural network models, if the same training samples are used to train the motion behavior prediction model and the ensemble learning model, and the same test samples are used to test the motion behavior prediction model and the ensemble learning model, the ensemble learning model may be over-fitted in the case of insufficient samples. To solve this problem, the embodiments of the present disclosure use the K-fold cross-validation method to train the motion behavior prediction model and the ensemble learning model. The K-fold cross-validation method is to divide the training set in the sample data set into K subsets.

[0175] After obtaining the K subsets, K-1 subsets are used as the training sample set to train the motion behavior prediction model, and the remaining one subset is used as the test set to train the motion behavior prediction model in turn. In the foregoing training process, each training sample set is used to train the model in turn to obtain the training prediction values P11, P12, …, P1K; after obtaining the foregoing prediction values, the foregoing training prediction values are averaged to obtain P1; in the foregoing training process, the intermediate motion behavior prediction model is used to process the training set in the sample data set to obtain the corresponding test prediction values T11, T12, …, T1K, and the foregoing test prediction values are averaged to obtain T1.

[0176] Since the rule-based operation method does not need to be trained, the training set P2 and the test set T2 used for training the ensemble learning model can be directly obtained.

[0177] Finally, [P1, P2] is used as the training set for training the ensemble learning model, and [T1, T2] is used as the test set for testing the ensemble learning model, to realize the training and testing of the ensemble learning model.

[0178] Of course, in other embodiments, in the case of sufficient test samples, the foregoing K-fold cross-validation method can not be used for model training, but the test samples can be directly divided into subsets for testing the motion behavior prediction model and the ensemble learning model, and the corresponding model is trained by using the subsets.

[0179] In addition to providing the foregoing method for predicting the motion behavior of the road participant, the embodiments of the present disclosure also provide a motion behavior prediction device. Figure 7 is a structural schematic diagram of the motion behavior prediction device provided by the embodiments of the present disclosure. As shown in Figure 7As shown, the motion behavior prediction device 700 provided by the embodiments of the present disclosure includes a data acquisition unit 701, a first prediction unit 702, a second prediction unit 703, a fusion prediction unit 704, and a motion behavior determination unit 705.

[0180] The data acquisition unit 701 is configured to acquire environmental perception data collected in M adjacent periods, where M≥2.

[0181] The first prediction unit 702 is configured to predict, according to at least part of the environmental perception data, first prediction probabilities of various motion behaviors of a target road participant by using a motion behavior prediction model, where the motion behavior prediction model is a pre-trained recurrent neural network model, and the motion behavior is a trend behavior of the target road participant relative to a reference object in a target lane.

[0182] The second prediction unit 703 is configured to predict, according to at least part of the environmental perception data, second prediction probabilities of various motion behaviors of the target road participant by using a rule-based prediction method.

[0183] The fusion prediction unit 704 is configured to perform data fusion on the first prediction probabilities and the second prediction probabilities corresponding to the various motion behaviors, to obtain third prediction probabilities of the various motion behaviors of the target road participant.

[0184] The motion behavior determination unit 705 is configured to determine a predicted motion behavior of the target road participant according to the third prediction probabilities corresponding to the various motion behaviors.

[0185] In some embodiments, the first prediction unit 702 performs data fusion on the first prediction probabilities and the second prediction probabilities corresponding to the various motion behaviors by using an ensemble learning model, where the ensemble learning model is a pre-trained neural network model.

[0186] In some embodiments, the environmental perception data includes first motion feature data of the target road participant, second motion feature data of a related road participant, and a position of a reference object, where the related road participant is a road participant directly affecting the motion behavior of the target road participant.

[0187] The first prediction unit 702 predicts, according to at least part of the first motion feature data in the M adjacent periods, at least part of the second motion feature data in the M adjacent periods, and the position of the reference object, the first prediction probabilities of the various motion behaviors of the target road participant by using the motion behavior prediction model.

[0188] In some embodiments, the first motion feature data comprises a first longitudinal velocity, a first longitudinal acceleration, a first lateral velocity and a first lateral position, the second motion feature data comprises a second longitudinal velocity and a second longitudinal acceleration, the position of the reference object comprises a lateral position of the reference object, and the first prediction unit 702 comprises a distance calculation subunit and a first probability prediction subunit.

[0189] The distance calculation subunit is configured to calculate a first lateral distance in each adjacent period according to the first lateral position and the lateral position of the reference object in the adjacent period.

[0190] The first probability prediction subunit is configured to process the first longitudinal velocity, the second longitudinal velocity, the first longitudinal acceleration, the second longitudinal acceleration, the first lateral velocity and the corresponding first lateral distance in the M adjacent periods by using a motion behavior prediction model to obtain the first prediction probability.

[0191] In some embodiments, the motion behavior prediction model comprises a first input fully connected network, a second input fully connected network, a recurrent neural network and an output fully connected network, and the first probability prediction subunit determines the first prediction probability by: processing the first longitudinal velocity, the second longitudinal velocity, the first longitudinal acceleration and the second longitudinal acceleration corresponding to each adjacent period by using the first input fully connected network to obtain a first intermediate vector corresponding to each adjacent period; processing the first lateral velocity and the corresponding first lateral distance in each adjacent period by using the second input fully connected network to obtain a second intermediate vector corresponding to each adjacent period; splicing the first intermediate vector and the second intermediate vector corresponding to each adjacent period to obtain a spliced vector; processing the spliced vector corresponding to each adjacent period in turn by using the recurrent neural network according to the period order of the adjacent periods to obtain a third intermediate vector; processing the third intermediate vector by using the output fully connected network to obtain an output vector, and determining the first prediction probability of the target road participant performing various motion behaviors based on the output vector.

[0192] In some embodiments, the environment perception data comprises the first motion feature data of the target road participant and the lateral position of the reference object, and the second prediction unit 703 obtains a second prediction probability of the target road participant performing various motion behaviors in the current period by using a rule-based prediction method according to the first motion feature data and the lateral position of the reference object in at least part of the M adjacent periods.

[0193] In some embodiments, the first motion feature data comprises a first lateral position and a first lateral velocity, and the second prediction unit 703 comprises a position determination subunit, a data lookup subunit, a comparison subunit, a behavior determination subunit and a second probability prediction subunit.

[0194] The position determining subunit is configured to determine a relative position of the target road participant according to the first lateral position corresponding to the most recent adjacent period and the lateral position of the reference object, the relative position being a position of the target road participant relative to the target lane.

[0195] The data searching subunit is configured to obtain the target confidence and the corresponding target lateral speed threshold in the lookup table.

[0196] The comparing subunit is configured to compare the first lateral speed corresponding to the most recent adjacent period and the target lateral speed threshold to obtain a first comparison result.

[0197] The behavior determining subunit is configured to determine the first motion behavior corresponding to the target confidence according to the first comparison result, the direction of the first lateral speed and the relative position.

[0198] The second probability predicting subunit is configured to take the target confidence as a second prediction probability corresponding to the first motion behavior, and determine second prediction probabilities corresponding to other motion behaviors according to the second prediction probability corresponding to the first motion behavior.

[0199] In some embodiments, the second probability predicting subunit determines a residual probability according to the second prediction probability corresponding to the first motion behavior, and determines the second prediction probabilities corresponding to the other motion behaviors according to the number of the other motion behaviors except the first motion behavior and the residual probability.

[0200] In some embodiments, the behavior determining subunit determines the first motion behavior by the following steps: in the case that the first lateral speed is less than the target lateral speed threshold, determining that the first motion behavior corresponding to the target confidence is parallel straight driving; in the case that the first lateral speed is greater than or equal to the target lateral speed threshold, the relative position is located in the target lane, and the direction of the first lateral speed points away from the target lane, determining that the motion behavior corresponding to the target confidence is lane-changing cut-out; in the case that the first lateral speed is greater than or equal to the target lateral speed threshold, the relative position is located in the adjacent lane of the target lane, and the direction of the first lateral speed points to the target lane, determining that the motion behavior corresponding to the target confidence is lane-changing cut-in.

[0201] In some embodiments, the target confidence in the lookup table corresponds to a plurality of lateral speed thresholds; the second predicting unit 703 further comprises a time length comparing subunit, a lateral distance calculating subunit and a lateral distance change value calculating subunit.

[0202] The time length comparing subunit is configured to compare the time length of the M adjacent periods and the first preset time length to obtain a second comparison result; in the case that the second comparison result is that the time length of the M adjacent periods is less than the first preset time length, taking a first set threshold in the plurality of lateral speed thresholds as the target lateral speed threshold.

[0203] The lateral distance calculation sub-unit is configured to calculate a corresponding lateral distance according to the lateral positions of the N most recent adjacent periods and the lateral position of the reference object, when the duration of the M adjacent periods is greater than or equal to the first preset duration.

[0204] The lateral distance change value calculation sub-unit is configured to obtain N-1 lateral distance change values according to the lateral distances of adjacent periods in the N most recent adjacent periods.

[0205] The data searching sub-unit is configured to take a second preset threshold value in the plurality of lateral speed threshold values as a target lateral speed threshold value, when the signs of the N-1 distance change values are the same, the second preset threshold value being smaller than the first preset threshold value; or take a third preset threshold value as the target lateral speed threshold value, when the signs of the N-P-1 distance change values corresponding to the N-P most recent adjacent periods are the same but the signs of the N-1 distance change values are not the same, P≤N-3, the third preset threshold value being greater than the second preset threshold value but smaller than the first preset threshold value.

[0206] In some embodiments, the first motion feature data further includes a first longitudinal distance of the target road participant relative to a target vehicle, the target vehicle being located in a lane where the reference object is located and adjacent to the target road participant; and the data searching sub-unit is configured to search the lookup table according to the first longitudinal distance to obtain the target confidence and the target lateral speed threshold value associated with the first longitudinal distance.

[0207] In some embodiments, the second prediction unit 703 further includes a road type obtaining sub-unit configured to obtain a road type of a road on which the target road participant travels; and the data searching sub-unit is configured to search the lookup table according to the road type to obtain the target confidence and the target lateral speed threshold value associated with the road type.

[0208] The data searching sub-unit is configured to search the lookup table according to the road type to obtain the target confidence and the target lateral speed threshold value associated with the road type.

[0209] In some embodiments, the second prediction unit 703 further includes a participant type determining sub-unit configured to determine a participant type of the target road participant, the participant type being a vehicle or a non-vehicle; and the data searching sub-unit is configured to search the lookup table according to the participant type to obtain the target confidence and the lateral speed threshold value associated with the participant type.

[0210] The disclosure also provides a computing device including a processor and a memory, wherein the memory stores a computer program, and when the computer program is executed by the processor, the motion behavior prediction method of any of the above embodiments can be implemented.

[0211] Figure 8is a structural schematic diagram of a computing device provided by embodiments of the present disclosure. The following will be specifically described with reference to Figure 8 which shows a structural schematic diagram suitable for implementing the computing device 800 in embodiments of the present disclosure. Figure 8 The computing device shown is merely an example and should not bring any limitation to the functions and use range of embodiments of the present disclosure.

[0212] As shown in Figure 8 , the computing device 800 can include a processing device (for example, a central processor, a graphics processor, etc.) 801, which can perform various appropriate actions and processes according to programs stored in a read-only memory (ROM) 802 or programs loaded from a storage device 808 into a random access memory (RAM) 803. In the RAM 803, various programs and data required for the operation of the computing device 800 are also stored. The processing device 801, the ROM 802, and the RAM 803 are connected to each other through a bus 804. An input / output (I / O) interface 805 is also connected to the bus 804.

[0213] Generally, the following devices can be connected to the I / O interface 805: input devices 806 including, for example, a touch screen, a touch pad, a camera, a microphone, an accelerometer, a gyroscope, etc.; output devices 807 including, for example, a liquid crystal display (LCD), a speaker, a vibrator, etc.; storage devices 808 including, for example, a magnetic tape, a hard disk, etc.; and communication devices 809. The communication devices 809 can allow the computing device 800 to communicate with other devices wirelessly or by wire to exchange data. Although Figure 8 The computing device 800 with various devices is shown, but it should be understood that it is not required to implement or have all the devices shown. More or fewer devices can be alternatively implemented or provided.

[0214] In particular, according to embodiments of the present disclosure, the processes described above with reference to the flowcharts can be implemented as a computer software program. For example, embodiments of the present disclosure include a computer program product comprising a computer program carried on a non-transitory computer readable medium, the computer program containing program code for executing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network through the communication devices 809, or installed from the storage devices 808, or installed from the ROM 802. When the computer program is executed by the processing device 801, the above-mentioned functions defined in the methods of embodiments of the present disclosure are performed.

[0215] It is noted that the aforementioned computer-readable medium of the present disclosure can be a computer-readable signal medium or a computer-readable storage medium or any combination thereof. The computer-readable storage medium can be, for example and without limitation, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any suitable combination of the foregoing. More specific examples of the computer-readable storage medium can include, but are not limited to, an electrical connection having one or more wires, a portable computer diskette, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or Flash memory), an optical fiber, a portable compact disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing. In the present disclosure, the computer-readable storage medium can be any tangible medium that contains or stores a program used by or in connection with an instruction execution system, apparatus, or device. In the present disclosure, the computer-readable signal medium can include a computer-readable program code transmitted by a computer-readable medium or a carrier wave transmits, propagates, or transfers a program used by or in connection with an instruction execution system, apparatus, or device. The program code contained in the computer-readable medium can be transmitted using any suitable medium, including but not limited to wire, cable, optical fiber, RF (radio frequency), or the like, or any suitable combination of the foregoing.

[0216] In some embodiments, the client, server, or both can communicate using any current known or future developed network protocol, such as HTTP (HyperText Transfer Protocol), and can be interconnected with any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include local area networks ("LAN"), wide area networks ("WAN"), internetworks (e.g., the Internet), and peer-to-peer networks (e.g., ad hoc peer-to-peer networks), as well as any current known or future developed networks.

[0217] The aforementioned computer-readable medium can be contained in the aforementioned computing device; or can exist separately from the computing device without being incorporated into the computing device.

[0218] The computer readable medium described above carries one or more programs, when the one or more programs are executed by the computing device, cause the computing device to: acquire M pieces of environment perception data collected in adjacent periods, M > 2; predict, according to at least part of the environment perception data, a first prediction probability of a target road participant performing various motion behaviors by using a motion behavior prediction model, the motion behavior prediction model being a pre-trained recurrent neural network model, and the motion behavior being a trend behavior of the target road participant relative to a reference object in a target lane; predict, according to at least part of the environment perception data, a second prediction probability of the target road participant performing various motion behaviors by using a rule-based prediction method; perform data fusion on the first prediction probability and the second prediction probability corresponding to various motion behaviors to obtain a third prediction probability of the target road participant performing various motion behaviors; and determine a predicted motion behavior of the target road participant according to the third prediction probability corresponding to various motion behaviors.

[0219] Computer program code for carrying out operations of the present disclosure can be written in any of one or more programming languages or combinations of languages including object or visual programming languages such as Java, Smalltalk, C++ or conventional procedural programming languages such as the "C" programming language or similar programming languages. The program code can execute entirely on the user's computer, partly on the user's computer, as a stand-alone software package, partly on the user's computer and partly on a remote computer or entirely on the remote computer or server. In the latter scenario, the remote computer can be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or the connection can be made to an external computer (for example, through the Internet using an Internet Service Provider).

[0220] The flow and block diagrams in the drawings show architectural, functional, and operational architectures of possible implementations of systems, methods, and computer program products according to various embodiments of the present disclosure. In this regard, each block in the flow and block diagrams can represent a module, a segment, or a portion of code, which comprises one or more executable instructions for implementing the specified logical function(s). It should also be noted that in some alternative implementations, the functions noted in the blocks can occur out of the order noted in the figures. For example, two blocks shown in succession may, in fact, be executed substantially concurrently or the blocks may be executed in the reverse order, depending on the functionality involved. It will also be noted that each block of the block diagrams and / or flowchart illustrations, and combinations thereof, can be implemented by special purpose hardware-based systems that perform the specified functions or operations, or combinations of special purpose hardware and computer instructions.

[0221] The units described in the embodiments of the present disclosure can be implemented by software, or can be implemented by hardware. In some cases, the names of the units do not constitute a limitation on the units themselves.

[0222] The functions described above in this document can be performed, at least in part, by one or more hardware logic components. For example, and without limitation, illustrative types of hardware logic components that can be used include Field-programmable Gate Arrays (FPGAs), Application-specific Integrated Circuits (ASICs), Application-specific Standard Products (ASSPs), System-on-a-chip systems (SOCs), Complex Programmable Logic Devices (CPLDs), etc.

[0223] In the context of the present disclosure, an automatically readable medium can be a tangible medium that can contain or store a program for use by or in connection with an instruction execution system, apparatus, or device. An automatically readable medium can be an automatically readable signal medium or an automatically readable storage medium. Automatically readable storage media can include, but are not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any suitable combination of the foregoing. More specific examples of automatically readable storage media can include electrical connections based on one or more wires, a portable computer diskette, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or Flash memory), an optical fiber, a portable compact disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0224] The embodiments of the present disclosure further provide a computer readable storage medium, the computer readable storage medium stores a computer program, when the computer program is executed by a processor, the method of any one of the above-mentioned method embodiments can be implemented, and the execution manner and beneficial effects are similar, which will not be described here.

[0225] It should be noted that, in this document, relational terms such as "first" and "second", and the like can be used solely to distinguish one entity or action from another entity or action without necessarily requiring or implying any actual such relationship or order between such entities or actions. Moreover, the terms "comprises", "comprising", or any other variations thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but can include other elements not expressly listed or inherent to such process, method, article, or apparatus. Without more limitations, an element defined by the statement "comprising a" does not exclude the existence of additional identical elements in the process, method, article, or apparatus that includes the element.

[0226] The foregoing is merely illustrative of the various ways and specific embodiments in which the disclosure can be carried out. Numerous modifications can be made to these embodiments without departing from the spirit and scope of the disclosure. Therefore, the disclosure is not limited to the specific embodiments described herein, but rather the scope of the disclosure is to be accorded the broadest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for predicting the motion behavior of road participants, characterized in that, include: Acquire environmental perception data collected in M ​​adjacent periods, where M≥2; the environmental perception data includes first motion characteristic data of the target road participant, second motion characteristic data of related road participants, and the position of a reference object, wherein the related road participants are road participants that directly affect the motion behavior of the target road participant; the first motion characteristic data includes first longitudinal velocity, first longitudinal acceleration, first lateral velocity, and first lateral position; the second motion characteristic data includes second longitudinal velocity and second longitudinal acceleration; the position of the reference object includes the lateral position of the reference object; Based on at least a portion of the environmental perception data, a motion behavior prediction model is used to predict the first predicted probability of a target road participant performing various motion behaviors, wherein the motion behavior prediction model is a pre-trained recurrent neural network model, and the motion behavior is the trend behavior of the target road participant relative to the target lane where the reference object is located. Based on at least a portion of the environmental perception data, a rule-based prediction method is used to predict a second predicted probability of the target road participant performing various motion behaviors; wherein the rule-based prediction method is based on a lookup table, which is used to determine different lateral velocity thresholds and corresponding target confidence levels; The first and second predicted probabilities corresponding to various motion behaviors are fused to obtain the third predicted probability of the target road participant performing various motion behaviors. The predicted motion behavior of the target road participant is determined based on the third prediction probability corresponding to various motion behaviors; The step of predicting the first probability of a target road participant performing various motion behaviors using a motion behavior prediction model based on at least a portion of the environmental perception data includes: Based on at least a portion of the first motion feature data in the M neighboring periods, at least a portion of the second motion feature data in the M neighboring periods, and the position of the reference object, the motion behavior prediction model is used to predict the first predicted probability of the target road participant performing various motion behaviors; The step of predicting the first probability of the target road participant performing various motion behaviors using the motion behavior prediction model based on at least a portion of the first motion feature data in the M neighboring periods, at least a portion of the second motion feature data in the M neighboring periods, and the position of the reference object includes: Calculate the first lateral distance in each adjacent period based on the first lateral position and the lateral position of the reference object in each adjacent period; The first predicted probability is obtained by processing the first longitudinal velocity, second longitudinal velocity, first longitudinal acceleration, second longitudinal acceleration, first lateral velocity, and corresponding first lateral distance in the M adjacent cycles using the motion behavior prediction model.

2. The method according to claim 1, characterized in that, The data fusion of the first and second predicted probabilities corresponding to various motion behaviors includes: An ensemble learning model is used to fuse the first and second predicted probabilities corresponding to various motion behaviors, wherein the ensemble learning model is a pre-trained neural network model.

3. The method according to claim 1, characterized in that, The motion behavior prediction model includes a first input fully connected network, a second input fully connected network, a recurrent neural network, and an output fully connected network. The step of processing the motion behavior prediction model to obtain the first predicted probability based on the first longitudinal velocity, second longitudinal velocity, first longitudinal acceleration, second longitudinal acceleration, first lateral velocity, and corresponding first lateral distance in the M adjacent cycles includes: The first input fully connected network is used to process the first longitudinal velocity, second longitudinal velocity, first longitudinal acceleration and second longitudinal acceleration corresponding to each adjacent period to obtain the first intermediate vector corresponding to each adjacent period; The second input fully connected network is used to process the first lateral velocity and the corresponding first lateral distance in each adjacent period to obtain the second intermediate vector corresponding to each adjacent period; By concatenating the first intermediate vector and the second intermediate vector corresponding to each adjacent period, a concatenated vector is obtained; According to the periodicity of the adjacent periods, the recurrent neural network is used to process the concatenated vectors corresponding to each adjacent period in turn to obtain the third intermediate vector. The third intermediate vector is processed by the fully connected output network to obtain an output vector, and the first predicted probability of the target road participant performing various motion behaviors is determined based on the output vector.

4. The method according to claim 1, characterized in that, The environmental perception data includes the first motion characteristic data of the target road participant and the lateral position of the reference object; The step of predicting a second probability of the target road participant performing various motion behaviors based on at least a portion of the environmental perception data using a rule-based prediction method includes: Based on the first motion feature data from at least a portion of the M neighboring periods and the lateral position of the reference object, a rule-based prediction method is used to obtain the second predicted probability of the target road participant performing various motion behaviors in the current period.

5. The method according to claim 4, characterized in that, The first motion feature data includes a first lateral position and a first lateral velocity; The step of obtaining the second predicted probability of the target road participant performing various motion behaviors in the current period using a rule-based prediction method based on the first motion feature data in at least a portion of the M neighboring periods and the lateral position of the reference object includes: The relative position of the target road participant is determined based on the first lateral position corresponding to the most recent cycle and the lateral position of the reference object, wherein the relative position is the position of the target road participant relative to the target lane; Retrieve the target confidence score and the corresponding target lateral velocity threshold from the lookup table; The first comparison result is obtained by comparing the magnitude of the first lateral velocity corresponding to the nearest period with the target lateral velocity threshold. Based on the first comparison result, the direction of the first lateral velocity, and the relative position, determine the first motion behavior corresponding to the target confidence level; The target confidence level is used as the second predicted probability corresponding to the first motion behavior; Based on the second predicted probability corresponding to the first motion behavior, determine the second predicted probability corresponding to other motion behaviors.

6. The method according to claim 5, characterized in that, The step of determining the second predicted probability corresponding to other motion behaviors based on the second predicted probability corresponding to the first motion behavior includes: The remaining probability is determined based on the second predicted probability corresponding to the first movement behavior; Based on the number of other movement behaviors besides the first movement behavior and the remaining probability, a second predicted probability corresponding to the other movement behaviors is determined.

7. The method according to claim 5, characterized in that, Determining the first motion behavior corresponding to the confidence level based on the first comparison result, the direction of the first lateral velocity, and the relative position includes: If the first lateral velocity is less than the target lateral velocity threshold, the first motion behavior corresponding to the confidence level is determined to be parallel straight movement; If the first lateral speed is greater than or equal to the target lateral speed threshold, the relative position is in the target lane, and the direction of the first lateral speed points away from the target lane, the motion behavior corresponding to the confidence level is determined to be lane change cut-out. If the first lateral speed is greater than or equal to the target lateral speed threshold, the relative position is in an adjacent lane of the target lane, and the direction of the first lateral speed is towards the target lane, then the motion behavior corresponding to the confidence level is determined to be lane change / cut-in.

8. The method according to claim 5, characterized in that, The target confidence level in the lookup table corresponds to multiple lateral velocity thresholds; Before comparing the magnitude of the first lateral velocity corresponding to the nearest neighbor period and the target lateral velocity threshold to obtain the first comparison result, the method further includes: The durations of the M adjacent cycles are compared with the first preset duration to obtain a second comparison result; if the second comparison result is that the durations of the M adjacent cycles are less than the first preset duration, the first preset threshold among the plurality of lateral velocity thresholds is used as the target lateral velocity threshold. If the first comparison result is that the duration of the M adjacent cycles is greater than or equal to the first preset duration, the corresponding lateral distance M≥N≥3 is calculated based on the first lateral position of the most recent N adjacent cycles and the lateral position of the reference object. Based on the lateral distances of the adjacent periods in the most recent N nearest periods, N-1 lateral distance change values ​​are obtained; If the signs of the N-1 distance change values ​​are all the same, the second preset threshold among the plurality of lateral velocity thresholds is used as the target lateral velocity threshold, and the second preset threshold is less than the first preset threshold; or... If the signs of the NP-1 distance change values ​​corresponding to the most recent NP adjacent cycles are all the same, but the signs of the N-1 distance change values ​​are different, then the third set threshold is used as the target lateral velocity threshold, P≤N-3, and the third set threshold is greater than the second set threshold but less than the first set threshold.

9. The method according to any one of claims 5-8, characterized in that, The first motion feature data also includes a first longitudinal distance between the target road participant and the target vehicle, wherein the target vehicle is located in the lane where the reference point is located and is close to the target road participant; The step of obtaining the target confidence score and the corresponding target lateral velocity threshold from the lookup table includes: The lookup table is retrieved based on the first longitudinal distance to obtain the target confidence and the target lateral velocity threshold associated with the first longitudinal distance.

10. The method according to any one of claims 5-8, characterized in that, The method further includes: Obtain the road type of the road on which the target road participant is driving; The step of obtaining the target confidence score and the corresponding target lateral velocity threshold from the lookup table includes: The lookup table is retrieved based on the road type to obtain the target confidence score and the target lateral speed threshold associated with the road type.

11. The method according to any one of claims 5-8, characterized in that, The method further includes: Determine the participant type of the target road participant, wherein the type is vehicle or non-vehicle; The step of obtaining the target confidence score and the corresponding target lateral velocity threshold from the lookup table includes: The lookup table is retrieved based on the participant type to obtain the target confidence score and the target lateral velocity threshold associated with the participant type.

12. A device for predicting the motion behavior of road participants, characterized in that, include: The data acquisition unit is used to acquire environmental perception data collected in M ​​adjacent periods, where M≥2; the environmental perception data includes first motion characteristic data of the target road participant, second motion characteristic data of related road participants, and the position of a reference object, wherein the related road participants are road participants that directly affect the motion behavior of the target road participant; the first motion characteristic data includes first longitudinal velocity, first longitudinal acceleration, first lateral velocity, and first lateral position; the second motion characteristic data includes second longitudinal velocity and second longitudinal acceleration; the position of the reference object includes the lateral position of the reference object; The first prediction unit is used to predict the first probability of a target road participant performing various motion behaviors based on at least a portion of the environmental perception data using a motion behavior prediction model. The motion behavior prediction model is a pre-trained recurrent neural network model, and the motion behavior is the trend behavior of the target road participant relative to the target lane where the reference object is located. The second prediction unit is used to predict the second prediction probability of the target road participant performing various motion behaviors based on at least a portion of the environmental perception data using a rule-based prediction method; wherein the rule-based prediction method is based on a lookup table, and the lookup table is used to determine the target confidence level corresponding to different lateral speed thresholds. The fusion prediction unit is used to fuse the first and second prediction probabilities corresponding to various motion behaviors to obtain the third prediction probability of the target road participant performing various motion behaviors. A motion behavior determination unit is used to determine the predicted motion behavior of the target road participant based on the third prediction probability corresponding to various motion behaviors. The step of predicting the first probability of a target road participant performing various motion behaviors using a motion behavior prediction model based on at least a portion of the environmental perception data includes: Based on at least a portion of the first motion feature data in the M neighboring periods, at least a portion of the second motion feature data in the M neighboring periods, and the position of the reference object, the motion behavior prediction model is used to predict the first predicted probability of the target road participant performing various motion behaviors; The first prediction unit includes: The distance calculation subunit is used to calculate the first lateral distance in each adjacent period based on the first lateral position and the lateral position of the reference object in each adjacent period; The first probability prediction subunit is used to process the first longitudinal velocity, second longitudinal velocity, first longitudinal acceleration, second longitudinal acceleration, first lateral velocity and corresponding first lateral distance in the M adjacent cycles using the motion behavior prediction model to obtain the first prediction probability.

13. The apparatus according to claim 12, characterized in that, The first prediction unit uses an ensemble learning model to fuse the first and second prediction probabilities corresponding to various motion behaviors, wherein the ensemble learning model is a pre-trained neural network model.

14. The apparatus according to claim 12, characterized in that, The motion behavior prediction model includes a first fully connected input network, a second fully connected input network, a recurrent neural network, and an output fully connected network; the first probability prediction subunit determines the first prediction probability according to the following steps: The first input fully connected network is used to process the first longitudinal velocity, second longitudinal velocity, first longitudinal acceleration and second longitudinal acceleration corresponding to each adjacent period to obtain the first intermediate vector corresponding to each adjacent period; The second input fully connected network is used to process the first lateral velocity and the corresponding first lateral distance in each adjacent period to obtain the second intermediate vector corresponding to each adjacent period; By concatenating the first intermediate vector and the second intermediate vector corresponding to each adjacent period, a concatenated vector is obtained; According to the periodicity of the adjacent periods, the recurrent neural network is used to process the concatenated vectors corresponding to each adjacent period in turn to obtain the third intermediate vector. The third intermediate vector is processed by the fully connected output network to obtain an output vector, and the first predicted probability of the target road participant performing various motion behaviors is determined based on the output vector.

15. The apparatus according to claim 12, characterized in that, The environmental perception data includes the first motion characteristic data of the target road participants and the lateral position of reference objects; The second prediction unit uses a rule-based prediction method to obtain the second predicted probability that the target road participant will perform various motion behaviors in the current period, based on the first motion feature data in at least some of the M neighboring periods and the lateral position of the reference object.

16. The apparatus according to claim 15, characterized in that, The first motion feature data includes a first lateral position and a first lateral velocity; the second prediction unit includes: The position determination subunit is used to determine the relative position of the target road participant based on the first lateral position corresponding to the nearest neighbor period and the lateral position of the reference object, wherein the relative position is the position of the target road participant relative to the target lane; The data lookup subunit is used to obtain the target confidence level and the corresponding target lateral velocity threshold from the lookup table; The comparison subunit is used to compare the magnitude of the first lateral velocity corresponding to the nearest neighbor period and the target lateral velocity threshold to obtain a first comparison result; The behavior determination subunit is used to determine the first motion behavior corresponding to the target confidence level based on the first comparison result, the direction of the first lateral velocity, and the relative position. The second probability prediction subunit is used to take the target confidence level as the second prediction probability corresponding to the first motion behavior, and to determine the second prediction probability corresponding to other motion behaviors based on the second prediction probability corresponding to the first motion behavior.

17. The apparatus according to claim 16, characterized in that, The second probability prediction subunit determines the remaining probability based on the second prediction probability corresponding to the first motion behavior, and determines the second prediction probability corresponding to other motion behaviors based on the number of other motion behaviors besides the first motion behavior and the remaining probability.

18. The apparatus according to claim 16, characterized in that, The behavior determination subunit determines the first motion behavior using the following steps: If the first lateral velocity is less than the target lateral velocity threshold, the first motion behavior corresponding to the confidence level is determined to be parallel straight movement; If the first lateral speed is greater than or equal to the target lateral speed threshold, the relative position is in the target lane, and the direction of the first lateral speed points away from the target lane, the motion behavior corresponding to the confidence level is determined to be lane change cut-out. If the first lateral speed is greater than or equal to the target lateral speed threshold, the relative position is in an adjacent lane of the target lane, and the direction of the first lateral speed is towards the target lane, then the motion behavior corresponding to the confidence level is determined to be lane change / cut-in.

19. The apparatus according to claim 16, characterized in that, The target confidence level in the lookup table corresponds to multiple lateral velocity thresholds; the second prediction unit further includes: The duration comparison subunit is used to compare the duration of the M adjacent cycles with the first preset duration to obtain a second comparison result; if the second comparison result is that the duration of the M adjacent cycles is less than the first preset duration, the first preset threshold among the plurality of lateral velocity thresholds is used as the target lateral velocity threshold. The lateral distance calculation subunit is used to calculate the corresponding lateral distance M≥N≥3 based on the first lateral position of the most recent N adjacent periods and the lateral position of the reference object when the first comparison result is that the duration of the M adjacent periods is greater than or equal to the first preset duration. The lateral distance change calculation subunit is used to obtain N-1 lateral distance change values ​​based on the lateral distances of adjacent periods in the most recent N nearest periods; When the signs of the N-1 distance change values ​​are all the same, the data lookup subunit uses the second set threshold among the multiple lateral velocity thresholds as the target lateral velocity threshold, where the second set threshold is less than the first set threshold; or, when the signs of the NP-1 distance change values ​​corresponding to the most recent NP adjacent cycles are all the same but the signs of the N-1 distance change values ​​are different, the third set threshold is used as the target lateral velocity threshold, where P≤N-3, and the third set threshold is greater than the second set threshold but less than the first set threshold.

20. The apparatus according to any one of claims 16-19, characterized in that, The first motion feature data also includes a first longitudinal distance between the target road participant and the target vehicle, wherein the target vehicle is located in the lane where the reference point is located and is close to the target road participant; The data lookup subunit retrieves the lookup table based on the first longitudinal distance to obtain the target confidence and the target lateral velocity threshold associated with the first longitudinal distance.

21. The apparatus according to any one of claims 16-19, characterized in that, The second prediction unit further includes: The road type acquisition subunit is used to acquire the road type of the road traveled by the target road participant; The data lookup subunit retrieves the lookup table based on the road type to obtain the target confidence level and the target lateral speed threshold associated with the road type.

22. The apparatus according to any one of claims 16-19, characterized in that, The second prediction unit further includes: The participant type determination subunit is used to determine the participant type of the target road participant, wherein the type is vehicle or non-vehicle; The data lookup subunit retrieves the lookup table based on the participant type to obtain the target confidence level and the target lateral velocity threshold associated with the participant type.

23. A computing device, characterized in that, Includes a processor and a memory, the memory being used to store computer programs; When the computer program is loaded by the processor, it causes the processor to execute the method for predicting the movement behavior of road participants as described in any one of claims 1-11.

24. A computer-readable storage medium, characterized in that, The storage medium stores a computer program that, when executed by a processor, causes the processor to implement the method for predicting the movement behavior of road participants as described in any one of claims 1-11.

Images