Vehicle driving trajectory prediction method, device, vehicle and storage medium

Abstract

The embodiment of the present invention discloses a vehicle driving trajectory prediction method, device, automobile and storage medium, which obtains the historical trajectory information of the target vehicle to be predicted and the surrounding vehicles, respectively determines the first trajectory feature and kinematic feature of the target vehicle to be predicted based on the historical trajectory information, and the second trajectory feature of the surrounding vehicles, determines the time attention weight of the first trajectory feature of the target vehicle to be predicted at the historical moment and the first trajectory feature at the current moment based on the first trajectory feature, the kinematic feature and the second trajectory feature, and determines the spatial attention weight of the surrounding vehicles and the target vehicle to be predicted; determines the spatiotemporal feature information according to the time attention weight and the spatial attention weight, and outputs the trajectory prediction result of the target vehicle to be predicted according to the spatiotemporal feature information through a decoder module. By considering the discrimination of the historical trajectory and the influence between vehicles, the accuracy of the vehicle driving trajectory prediction can be improved, thereby ensuring driving safety.

Description

Technical Field

[0001] The present invention relates to the field of vehicle driving technology, and in particular to a vehicle driving trajectory prediction method, device, vehicle and storage medium. Background Art

[0002] With the continuous development of social economy, the number of smart driving cars on the road is gradually increasing, and the road traffic situation is becoming more and more complicated. In a complex traffic environment, it is more difficult to predict the vehicle's driving trajectory.

[0003] At present, when faced with high-dynamic scenarios, smart driving cars are unable to accurately predict the driving behaviors and trajectories of other vehicles around them, resulting in reduced driving safety and the inability of the vehicle's decision-making and planning modules to work properly. Ordinary vehicle trajectory prediction schemes do not consider trajectory changes or driving behavior mutations caused by mutual influences between vehicles. At the same time, they lack the ability to distinguish the only historical trajectory input information, and data-driven prediction schemes face the problem of poor interpretability. Therefore, the accuracy of vehicle trajectories predicted using existing technologies is low. Summary of the invention

[0004] The present invention provides a vehicle driving trajectory prediction method, device, vehicle and storage medium to improve the accuracy of vehicle driving trajectory prediction, thereby ensuring driving safety.

[0005] According to one aspect of the present invention, a vehicle driving trajectory method is provided, comprising: obtaining historical trajectory information of a target vehicle to be predicted and surrounding vehicles, and determining a first trajectory feature and a kinematic feature of the target vehicle to be predicted, and a second trajectory feature of the surrounding vehicles based on the historical trajectory information;

[0006] Inputting the first trajectory feature and the kinematic feature into a time attention module, and determining the time attention weight of the first trajectory feature of the target vehicle to be predicted at a historical moment and the first trajectory feature at a current moment through the time attention module;

[0007] Inputting the first trajectory feature, the kinematic feature and the second trajectory feature into a spatial attention module, and determining the spatial attention weights of the surrounding vehicles and the target vehicle to be predicted by the spatial attention module;

[0008] The spatiotemporal feature information of the target vehicle to be predicted is determined according to the temporal attention weight and the spatial attention weight, and the trajectory prediction result of the target vehicle to be predicted is outputted according to the spatiotemporal feature information through a decoder module.

[0009] According to another aspect of the present invention, a vehicle driving trajectory device is provided, comprising: a historical trajectory information acquisition module, used to acquire historical trajectory information of a target vehicle to be predicted and surrounding vehicles, and determine a first trajectory feature and a kinematic feature of the target vehicle to be predicted, and a second trajectory feature of the surrounding vehicles based on the historical trajectory information;

[0010] a time attention weight determination module, configured to input the first trajectory feature and the kinematic feature into a time attention module, and determine the time attention weight of the first trajectory feature at a historical moment and the first trajectory feature at a current moment of the target vehicle to be predicted through the time attention module;

[0011] a spatial attention weight determination module, configured to input the first trajectory feature, the kinematic feature, and the second trajectory feature into a spatial attention module, and determine the spatial attention weights of the surrounding vehicles and the target vehicle to be predicted through the spatial attention module;

[0012] The trajectory prediction result output module is used to determine the spatiotemporal feature information of the target vehicle to be predicted according to the temporal attention weight and the spatial attention weight, and output the trajectory prediction result of the target vehicle to be predicted according to the spatiotemporal feature information through a decoder module.

[0013] According to another aspect of the present invention, there is provided a vehicle, the vehicle comprising:

[0014] at least one processor; and

[0015] a memory communicatively connected to the at least one processor; wherein,

[0016] The memory stores a computer program that can be executed by the at least one processor, and the computer program is executed by the at least one processor so that the at least one processor can execute a vehicle driving trajectory method described in any embodiment of the present invention.

[0017] According to another aspect of the present invention, a computer-readable storage medium is provided, wherein the computer-readable storage medium stores computer instructions, and the computer instructions are used to enable a processor to implement a vehicle driving trajectory method described in any embodiment of the present invention when executed.

[0018] The technical solution of the embodiment of the present invention can obtain the historical trajectory information of the target vehicle to be predicted and the surrounding vehicles, determine the first trajectory feature and kinematic feature of the target vehicle to be predicted, and the second trajectory feature of the surrounding vehicles based on the historical trajectory information, determine the time attention weight of the first trajectory feature of the target vehicle to be predicted at the historical moment and the first trajectory feature at the current moment based on the first trajectory feature, the kinematic feature and the second trajectory feature, and determine the spatial attention weight of the surrounding vehicles and the target vehicle to be predicted; determine the spatiotemporal feature information according to the time attention weight and the spatial attention weight, and output the trajectory prediction result of the target vehicle to be predicted according to the spatiotemporal feature information through the decoder module. In the process of vehicle driving trajectory prediction, the embodiment of the present invention combines the discrimination of historical trajectory information and the mutual influence between vehicles, which can improve the accuracy of vehicle driving trajectory prediction, thereby ensuring driving safety.

[0019] It should be understood that the contents described in this section are not intended to identify the key or important features of the embodiments of the present invention, nor are they intended to limit the scope of the present invention. Other features of the present invention will become easily understood through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0021] Figure 1 is a flow chart of a vehicle driving trajectory method provided according to Embodiment 1 of the present invention;

[0022] Figure 2 is a flow chart of a vehicle driving trajectory method provided according to Embodiment 2 of the present invention;

[0023] Figure 3 is a schematic diagram of the structure of a vehicle driving track device provided according to Embodiment 3 of the present invention;

[0024] Figure 4 It is a structural schematic diagram of a car that implements a vehicle driving trajectory method provided by Embodiment 4 of the present invention. DETAILED DESCRIPTION

[0025] In order to enable those skilled in the art to better understand the scheme of the present invention, the technical scheme in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present invention.

[0026] It should be noted that the terms "first", "second", etc. in the specification and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged where appropriate, so that the embodiments of the present invention described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or automobile that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or automobiles.

[0027] Embodiment 1

[0028] Figure 1 A flow chart of a vehicle driving trajectory method is provided for the first embodiment of the present invention. This embodiment is applicable to vehicle driving scenarios. The method can be executed by a vehicle driving trajectory device. The vehicle driving trajectory device can be implemented in the form of hardware and / or software. The vehicle driving trajectory device can be configured in a car, especially in a vehicle driving trajectory device configured in a car. Figure 1 As shown, the method includes:

[0029] S110, obtaining historical trajectory information of the target vehicle to be predicted and surrounding vehicles, and determining first trajectory characteristics and kinematic characteristics of the target vehicle to be predicted and second trajectory characteristics of the surrounding vehicles based on the historical trajectory information.

[0030] Among them, the target vehicle to be predicted can be a moving vehicle equipped with the vehicle driving trajectory prediction provided by an embodiment of the present invention; the surrounding vehicles can be one or more vehicles selected near the surrounding vehicles to be predicted after the target vehicle to be predicted is determined; the historical trajectory information can be the driving trajectory information generated by the vehicle driving along the lane during a certain period of time in the past. Furthermore, the vehicle driving trajectory information can be composed of a set of discrete two-dimensional coordinate points.

[0031] Specifically, the historical trajectory information of the target vehicle to be predicted and the historical trajectory information of the surrounding vehicles can be represented by the following formulas:

[0032]

[0033]

[0034] Among them, {X tar ,X nbs} represents the historical trajectory information of the target vehicle to be predicted and the historical trajectory information of the surrounding vehicles, t=1,2,…,T obs Represents the time span of the historical trajectory information that can be observed, Represents the specific longitudinal and lateral coordinate points of the vehicle in two-dimensional space; * represents the cross-correlation operation, which is specifically used to realize the convolution of historical trajectory information.

[0035] The kinematic characteristics may be vehicle motion characteristics that the vehicle should satisfy certain kinematic laws and constraints during driving.

[0036] Exemplarily, S110 may include embedding historical trajectory information of the target vehicle to be predicted and surrounding vehicles respectively through a nonlinear activation function, encoding the embedded historical trajectory information through a long short-term memory network to obtain a first trajectory feature of the target vehicle to be predicted and a second trajectory feature of the surrounding vehicles; and determining the kinematic features of the target vehicle to be predicted based on the historical trajectory information of the target vehicle to be predicted through a fully connected layer network.

[0037] The first trajectory information and the second trajectory information are encoded information obtained by encoding the embedded historical trajectory information based on the long short-term memory network. Furthermore, the embedding of the historical trajectory information of the target vehicle to be predicted and the surrounding vehicles can be achieved through the following formula:

[0038] e t =LeakyReLU(W e *X t +b e );

[0039] in, They represent the information of the target vehicle to be predicted and its surrounding vehicle trajectory information after being embedded through the nonlinear activation function. LeakyReLU(·) represents the nonlinear activation function. represents the historical trajectory of the target vehicle and its surrounding vehicles, {W e ,W b} represent the weights and biases of the embedding function to be trained.

[0040] Furthermore, the embedded historical trajectory information can be encoded through a long short-term memory network, and the encoded information at each moment can be stored in a database.

[0041] Specifically, the first trajectory feature of the target vehicle to be predicted and the second trajectory feature of the surrounding vehicles can be expressed by the following formula:

[0042] h t =LSTM(W l ·e t +b l );

[0043] in, can represent the first trajectory feature of the target vehicle to be predicted and the second trajectory feature of the surrounding vehicles respectively. LSTM represents the long short-term memory network module. l ,W l} respectively represent the weight and bias of LSTM to be trained;

[0044] Since the target vehicle to be predicted should satisfy certain kinematic laws and constraints, it is necessary to perform additional kinematic feature extraction on the target vehicle to be predicted. Specifically, a single-layer fully connected network (FCN) can be used to determine its kinematic features.

[0045] Furthermore, the following formula can be used to extract the kinematic characteristics of the target vehicle to be predicted:

[0046]

[0047] in, represents the kinematic mapping information of the target vehicle to be predicted, FC represents the fully connected layer network, and W f Represents the weight of the fully connected layer network.

[0048] S120, inputting the first trajectory feature and the kinematic feature into a time attention module, and determining the time attention weight of the first trajectory feature of the target vehicle to be predicted at a historical moment and the first trajectory feature at a current moment through the time attention module.

[0049] Among them, the time attention module can be to sample the importance of trajectory coordinates at different historical moments in vehicle driving trajectory prediction, and then filter out the trajectory information of historical moments that have a greater impact on the prediction results and increase the time attention weight, thereby deepening its importance.

[0050] Exemplarily, determining the time attention weights of the historical moment trajectories and the current moment trajectory of the target vehicle to be predicted through the time attention module may include: calculating the time correlations between the first trajectory features of the target vehicle to be predicted at the current moment and the first trajectory features of each historical moment through a similarity function; calculating the time attention weights of the first trajectory features of each historical moment and the first trajectory features of the current moment according to the time correlation through an activation function and a normalized exponential function, wherein the time attention module includes a similarity function, an activation function and a normalized exponential function.

[0051] Furthermore, for the target vehicle to be predicted, the first trajectory feature calculated at the current moment is The first trajectory characteristics of each moment in its history are In this embodiment, the importance of each historical node can be determined by comparing the first trajectory feature with the first trajectory information of each historical moment. Specifically, the importance of the historical node can be expressed by the following formula:

[0052]

[0053] Specifically, the time correlation between the first trajectory feature of the target vehicle to be predicted at the current moment and the first trajectory feature of each historical moment is calculated by the similarity function and can be expressed by the following formula:

[0054]

[0055] Where f(·) represents the correlation measurement function, W r Represents the linear transformation weights to be learned.

[0056] Furthermore, the time attention weights of each historical moment can be calculated according to the time correlation according to the following formula:

[0057]

[0058] Among them, α i It represents the proportion of each historical node in the current prediction, that is, the temporal attention weight, softmax(·) is the normalized exponential function, and tanh(·) is the hyperbolic tangent activation function.

[0059] S130, inputting the first trajectory feature, the kinematic feature and the second trajectory feature into a spatial attention module, and determining the spatial attention weights of the surrounding vehicles and the target vehicle to be predicted through the spatial attention module.

[0060] Among them, the spatial attention module can be used to describe the interaction information between vehicles during driving.

[0061] Specifically, assuming that the target vehicle to be predicted interacts with the surrounding n vehicles in space, the trajectory features corresponding to the surrounding n vehicles at each historical moment can be calculated according to S110-S120, which can be recorded as S, that is:

[0062]

[0063] Exemplarily, determining the spatial attention weights of surrounding vehicles and the target vehicle to be predicted through the spatial attention module may include: calculating the spatial correlation between the first trajectory feature and the second trajectory feature through a cosine distance measurement function; calculating the spatial attention weights of the surrounding vehicles and the target vehicle to be predicted according to the spatial correlation through a normalization function, wherein the spatial attention module includes a cosine distance measurement function and a normalization function.

[0064] Among them, the trajectory characteristics of each vehicle include the trajectory characteristic information at each historical moment, namely:

[0065]

[0066] Specifically, the spatial correlation between the first trajectory feature and the second trajectory feature calculated by the cosine distance metric function can be expressed by the following formula:

[0067]

[0068] Among them, g(·) is the cosine distance metric function;

[0069] Specifically, the spatial attention weights of the surrounding vehicles and the target vehicle to be predicted are calculated according to the spatial correlation through the normalization function and can be expressed by the following formula:

[0070]

[0071] Among them, α′ i is the spatial attention weight.

[0072] S140, determining the spatiotemporal feature information of the target vehicle to be predicted according to the temporal attention weight and the spatial attention weight, and outputting the trajectory prediction result of the target vehicle to be predicted according to the spatiotemporal feature information through a decoder module.

[0073] Among them, the spatiotemporal feature information is tensor information that has no actual physical meaning, which may include fitting information of vehicle trajectories, interaction information between vehicles, and importance information of trajectories; the decoder can be used to decode the spatiotemporal feature information and convert it into a prediction result of the trajectory of the target vehicle to be predicted.

[0074] Exemplarily, S140 may include: determining a predicted coordinate point based on a preconfigured prediction time domain and sampling frequency; for each predicted coordinate point, calculating a context vector containing temporal attention information according to the temporal attention weight, and calculating a context vector containing spatial attention information according to the spatial attention weight; determining the spatiotemporal feature information of the target vehicle to be predicted corresponding to the predicted coordinate point according to the context vector containing the temporal attention information and the context vector containing the spatial attention information of each predicted coordinate point through a fully connected layer network; and outputting the trajectory prediction result of the target vehicle to be predicted according to the spatiotemporal feature information of each predicted coordinate point through a decoder module.

[0075] Exemplarily, calculating a context vector containing temporal attention information according to the temporal attention weight, and calculating a context vector containing spatial attention information according to the spatial attention weight may include: calculating a context vector containing temporal attention information according to the first trajectory features and the temporal attention weights of each historical moment through the temporal attention module; and calculating a context vector containing spatial attention information according to the second trajectory features and the spatial attention weights of surrounding vehicles through the spatial attention module.

[0076] Specifically, the temporal attention module calculates the context vector containing the temporal attention information according to the first trajectory features and temporal attention weights of each historical moment, which can be expressed by the following formula:

[0077]

[0078] Among them, v t represents the context vector containing temporal attention information, and k represents the number of trajectory features contained in the trajectory of each historical moment, such as the preset time domain T pred is 5 seconds, the sampling frequency is 0.2 seconds, and k=25 is calculated at this time;

[0079] Specifically, the context vector containing the spatial attention information is calculated by the spatial attention module according to the second trajectory features of the surrounding vehicles and the spatial attention weights, which can be expressed by the following formula:

[0080]

[0081] Among them, v′ t represents the context vector containing temporal attention information,

[0082] Specifically, the spatiotemporal characteristic information of the target vehicle to be predicted corresponding to the prediction coordinate point can be expressed by the following formula:

[0083] h final =W v v t +W v′v′ t ;

[0084] Among them, h final It is the spatiotemporal feature information that includes temporal attention and spatial attention.

[0085] Specifically, the coordinate points of the future trajectory predicted by the LSTM decoder can be expressed as follows:

[0086] Y tar =LSTM(h final ; W final ,b fianl );

[0087] in,

[0088]

[0089] The embodiment of the present invention obtains the historical trajectory information of the target vehicle to be predicted and the surrounding vehicles, determines the first trajectory feature and kinematic feature of the target vehicle to be predicted and the second trajectory feature of the surrounding vehicles based on the historical trajectory information, determines the time attention weight of the first trajectory feature of the target vehicle to be predicted at the historical moment and the first trajectory feature at the current moment based on the first trajectory feature, kinematic feature and second trajectory feature, and determines the spatial attention weight of the surrounding vehicles and the target vehicle to be predicted; determines the spatiotemporal feature information according to the time attention weight and the spatial attention weight, and outputs the trajectory prediction result of the target vehicle to be predicted according to the spatiotemporal feature information through the decoder module. By applying different spatial weights to the surrounding vehicles, different spatial influences are reflected; and a time attention mechanism is designed to analyze the correlation between the current state and the historical state of the target vehicle to be predicted, and the time weights of different historical moments are calculated to reflect different time influences; the accuracy of vehicle driving trajectory prediction can be improved, thereby ensuring driving safety.

[0090] Embodiment 2

[0091] Figure 2 This is a flow chart of a vehicle driving trajectory method provided by Embodiment 2 of the present invention. Based on the above embodiments, this embodiment further optimizes "outputting the trajectory prediction result of the target vehicle to be predicted according to the spatiotemporal feature information of each prediction coordinate point through the decoder module". The explanations of the terms that are the same or corresponding to the above embodiments are not repeated here. Figure 3 The vehicle driving trajectory method provided in this embodiment specifically includes the following steps:

[0092] S210, obtaining historical trajectory information of the target vehicle to be predicted and surrounding vehicles, and determining first trajectory characteristics and kinematic characteristics of the target vehicle to be predicted and second trajectory characteristics of the surrounding vehicles based on the historical trajectory information.

[0093] S220, inputting the first trajectory feature and the kinematic feature into a time attention module, and determining the time attention weight of the first trajectory feature of the target vehicle to be predicted at a historical moment and the first trajectory feature at a current moment through the time attention module.

[0094] S230, inputting the first trajectory feature, the kinematic feature, and the second trajectory feature into a spatial attention module, and determining the spatial attention weights of the surrounding vehicles and the target vehicle to be predicted through the spatial attention module.

[0095] S240. Determine the spatiotemporal feature information of the target vehicle to be predicted according to the temporal attention weight and the spatial attention weight.

[0096] S250, mapping the spatiotemporal feature information of the target vehicle to be predicted at each prediction coordinate point to the driving behavior space through a decoder module to obtain a driving behavior prediction result of each prediction coordinate point; determining a driving trajectory of the target vehicle to be predicted according to the driving behavior prediction result of each prediction coordinate point;

[0097] The decoder module is constructed based on the following driving behaviors: longitudinal acceleration, constant speed, deceleration, lateral left lane change, lane keeping and right lane change.

[0098] Among them, driving behavior can be defined by the following formula:

[0099] M={m k |k=1,2,…6};

[0100] Driving behaviors can include longitudinal acceleration, constant speed and deceleration, lateral left lane change, lane keeping and right lane change. By mapping the spatiotemporal feature information of the target vehicle to be predicted at each prediction coordinate point to the driving behavior space, six driving behavior prediction results can be obtained.

[0101] Specifically, taking the predicted longitudinal driving behavior as an example, the predicted longitudinal driving behavior and the longitudinal driving behavior probability can be expressed by the following formula:

[0102] B lon =FC(h final ; W lon );

[0103] B lat =FC(h final ; W lat );

[0104] Among them, {B lon ,B lat} respectively represent the predicted longitudinal driving behavior and longitudinal driving behavior probability, {W lon ,Wlat} respectively represent the weights of the fully connected layer network for longitudinal driving behavior and lateral driving behavior;

[0105] Specifically, assuming that the predicted trajectory coordinates satisfy a bivariate Gaussian distribution, their mean can be expressed by the following formula:

[0106]

[0107] in, They represent the predicted longitudinal position and the predicted lateral position respectively, and their standard deviation can be expressed by the following formula:

[0108]

[0109] in, They represent the predicted longitudinal standard deviation and the predicted lateral standard respectively, and the correlation coefficient of the bivariate Gaussian distribution is ρ t ;

[0110] Consider the prediction time domain T pred is 5 seconds, at t∈(t+1,t+T pred ) moment, the prediction result should satisfy the following formula:

[0111]

[0112] in, Indicates that the target vehicle to be predicted is at (t+1, t+T pred ) is the predicted position at a certain moment in the interval, then the posterior probability distribution of the future trajectory of the target vehicle to be predicted can be expressed as follows:

[0113]

[0114] in, represents the conditional transition probability from the historical trajectory information of the target vehicle to be predicted and its surrounding vehicles to the predicted trajectory of the target vehicle, It represents the parameters of the binary Gaussian distribution at each moment in the prediction time domain. The posterior probability is decomposed into first calculating the probability of driving behavior through the historical trajectories of the target vehicle to be predicted and its surrounding vehicles, and then further predicting the future trajectory in combination with the driving behavior.

[0115] The above steps establish the probability transfer relationship between historical trajectory-driving behavior-predicted trajectory. The loss function can be constructed by the root mean square error (RMSE), which can be expressed by the following formula:

[0116]

[0117] S260. Determine the spatial attention weight distribution information according to the spatial attention weight of each predicted coordinate point, determine the temporal attention weight distribution information according to the temporal attention weight of each predicted coordinate point, and output the driving trajectory, driving behavior prediction results, spatial attention weight distribution information and temporal attention weight distribution information.

[0118] In this embodiment, the driving data can be continuously sent into the model in an iterative form for optimization, and the network parameters that minimize the overall loss function error can be obtained to complete the model training. When new data flows into the trained prediction model, the driving trajectory of the target vehicle to be predicted, the driving behavior prediction results, the time attention weight distribution and the space attention weight distribution are output. The first two reactions can reflect the accuracy of the prediction results, and the latter two can reflect the importance sampling of the input data and the size of the interaction between the target vehicle and the surrounding vehicles.

[0119] The embodiment of the present invention can also map the spatiotemporal feature information of the target vehicle to be predicted at each prediction coordinate point to the driving behavior space through the decoder module, obtain the driving behavior prediction results of each prediction coordinate point to predict the vehicle's driving trajectory, improve the accuracy and rationality of the prediction, and the output information includes not only the driving behavior and future trajectory, but also the spatial attention weight distribution and the temporal attention weight distribution, which is helpful to analyze the vehicle driving characteristics for different traffic scenarios, make the data-driven trajectory prediction model interpretable, and make up for the shortcomings of the traditional deep learning method that is abstract and complex.

[0120] Embodiment 3

[0121] Figure 3 This is a schematic diagram of the structure of a vehicle driving track device provided by Embodiment 3 of the present invention. Figure 3 As shown, the device comprises:

[0122] A historical trajectory information acquisition module 310 is used to acquire historical trajectory information of the target vehicle to be predicted and surrounding vehicles, and determine the first trajectory characteristics and kinematic characteristics of the target vehicle to be predicted and the second trajectory characteristics of the surrounding vehicles based on the historical trajectory information;

[0123] A time attention weight determination module 320, configured to input the first trajectory feature and the kinematic feature into a time attention module, and determine the time attention weight of the first trajectory feature at a historical moment and the first trajectory feature at a current moment of the target vehicle to be predicted through the time attention module;

[0124] A spatial attention weight determination module 330, configured to input the first trajectory feature, the kinematic feature, and the second trajectory feature into a spatial attention module, and determine the spatial attention weights of the surrounding vehicles and the target vehicle to be predicted through the spatial attention module;

[0125] The trajectory prediction result output module 340 is used to determine the spatiotemporal feature information of the target vehicle to be predicted according to the temporal attention weight and the spatial attention weight, and output the trajectory prediction result of the target vehicle to be predicted according to the spatiotemporal feature information through the decoder module.

[0126] Optionally, the historical trajectory information acquisition module 310 may include:

[0127] A historical trajectory information encoding unit is used to embed the historical trajectory information of the target vehicle to be predicted and the surrounding vehicles respectively through a nonlinear activation function, and encode the embedded historical trajectory information through a long short-term memory network to obtain a first trajectory feature of the target vehicle to be predicted and a second trajectory feature of the surrounding vehicles;

[0128] The kinematic feature determination unit is used to determine the kinematic features of the target vehicle to be predicted based on the historical trajectory information of the target vehicle to be predicted through a fully connected layer network.

[0129] Optionally, the temporal attention weight determination module 320 may include:

[0130] A first trajectory feature and time attention weight calculation unit, used in a similarity function calculation module, for respectively calculating the time correlation between the first trajectory feature of the target vehicle to be predicted at the current moment and the first trajectory features of each historical moment through a similarity function;

[0131] Optionally, the spatial attention weight determination module 330 may include:

[0132] A cosine distance metric function calculation unit, used to calculate the spatial correlation between the first trajectory feature and the second trajectory feature by using the cosine distance metric function;

[0133] A normalization function calculation unit is used to calculate the spatial attention weights of surrounding vehicles and the target vehicle to be predicted according to spatial correlation through a normalization function, wherein the spatial attention module includes a cosine distance measurement function and a normalization function.

[0134] Optionally, the trajectory prediction result output 340 may include:

[0135] A prediction coordinate point prediction unit, used to determine the prediction coordinate point based on a pre-configured prediction time domain and sampling frequency;

[0136] A context vector calculation unit, used to calculate, for each predicted coordinate point, a context vector containing temporal attention information according to the temporal attention weight, and a context vector containing spatial attention information according to the spatial attention weight;

[0137] A spatiotemporal feature information determination unit, configured to determine the spatiotemporal feature information of the target vehicle to be predicted corresponding to the predicted coordinate point according to the context vector containing the temporal attention information and the context vector containing the spatial attention information of each predicted coordinate point through a fully connected layer network;

[0138] The trajectory prediction result output unit is used to output the trajectory prediction result of the target vehicle to be predicted according to the spatiotemporal feature information of each prediction coordinate point through the decoder module.

[0139] Optionally, the context vector calculation unit may include:

[0140] A context vector calculation subunit of the temporal attention information, used to calculate a context vector containing the temporal attention information according to the first trajectory features and the temporal attention weights of each historical moment through the temporal attention module;

[0141] The context vector calculation subunit of the spatial attention information is used to calculate the context vector containing the spatial attention information according to the second trajectory characteristics of the surrounding vehicles and the spatial attention weights through the spatial attention module.

[0142] Optionally, the trajectory prediction result output module 340 may include:

[0143] A mapping unit, used to map the spatiotemporal feature information of the target vehicle to be predicted at each prediction coordinate point to the driving behavior space through a decoder module to obtain a driving behavior prediction result at each prediction coordinate point;

[0144] A driving trajectory prediction unit, used to determine the driving trajectory of the target vehicle to be predicted according to the driving behavior prediction results of each prediction coordinate point;

[0145] The spatial attention weight distribution information determination unit is used to determine the spatial attention weight distribution information according to the spatial attention weights of each predicted coordinate point, determine the temporal attention weight distribution information according to the temporal attention weights of each predicted coordinate point, and output the driving trajectory, driving behavior prediction results, spatial attention weight distribution information and temporal attention weight distribution information; wherein the decoder module is based on the following driving behavior constructions: longitudinal acceleration, constant speed, deceleration, lateral left lane change, lane keeping and right lane change.

[0146] A vehicle driving trajectory device provided by an embodiment of the present invention can execute a vehicle driving trajectory method provided by any embodiment of the present invention, and has functional modules and beneficial effects corresponding to the execution method.

[0147] Embodiment 4

[0148] Figure 4 This is a schematic diagram of the structure of a car provided by the fourth embodiment of the present invention. Figure 4 As shown, the automobile 10 includes at least one processor 11, and a memory connected to the at least one processor 11, such as a read-only memory (ROM) 12, a random access memory (RAM) 13, etc., wherein the memory stores a computer program that can be executed by at least one processor, and the processor 11 can perform various appropriate actions and processes according to the computer program stored in the read-only memory (ROM) 12 or the computer program loaded from the storage unit 18 to the random access memory (RAM) 13. In the RAM 13, various programs and data required for the operation of the automobile 10 can also be stored. The processor 11, the ROM 12, and the RAM 13 are connected to each other through a bus 14. An input / output (I / O) interface 15 is also connected to the bus 14.

[0149] The processor 11 may be a variety of general and / or special processing components with processing and computing capabilities. Some examples of the processor 11 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special artificial intelligence (AI) computing chips, various processors running machine learning model algorithms, a digital signal processor (DSP), and any appropriate processor, controller, microcontroller, etc. When the processor 11 executes the various methods described above, it implements:

[0150] Acquire historical trajectory information of the target vehicle to be predicted and surrounding vehicles, and determine first trajectory characteristics and kinematic characteristics of the target vehicle to be predicted and second trajectory characteristics of the surrounding vehicles based on the historical trajectory information;

[0151] The first trajectory feature and the kinematic feature are input into a time attention module, and the time attention weight of the first trajectory feature at a historical moment and the first trajectory feature at a current moment of the target vehicle to be predicted is determined by the time attention module;

[0152] The first trajectory feature, the kinematic feature and the second trajectory feature are input into a spatial attention module, and the spatial attention weights of the surrounding vehicles and the target vehicle to be predicted are determined by the spatial attention module;

[0153] The spatiotemporal feature information of the target vehicle to be predicted is determined according to the temporal attention weight and the spatial attention weight, and the trajectory prediction result of the target vehicle to be predicted is output through the decoder module according to the spatiotemporal feature information.

[0154] Computer programs for implementing the methods of the present invention may be written in any combination of one or more programming languages. These computer programs may be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device, so that when the computer program is executed by the processor, the functions / operations specified in the flow chart and / or block diagram are implemented. The computer program may be executed entirely on the machine, partially on the machine, partially on the machine and partially on a remote machine as a stand-alone software package, or entirely on a remote machine or server.

[0155] In the context of the present invention, a computer-readable storage medium may be a tangible medium that may contain or store a computer program for use by or in combination with an instruction execution system, device or equipment. A computer-readable storage medium may include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, devices or equipment, or any suitable combination of the foregoing. Alternatively, a computer-readable storage medium may be a machine-readable signal medium. A more specific example of a machine-readable storage medium may include an electrical connection based on one or more lines, a portable computer disk, 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 disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0156] It should be understood that the various forms of processes shown above can be used to reorder, add or delete steps. For example, the steps described in the present invention can be executed in parallel, sequentially or in different orders, as long as the desired results of the technical solution of the present invention can be achieved, and this document does not limit this.

[0157] The above specific implementations do not constitute a limitation on the protection scope of the present invention. It should be understood by those skilled in the art that various modifications, combinations, sub-combinations and substitutions can be made according to design requirements and other factors. Any modification, equivalent substitution and improvement made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A vehicle driving trajectory prediction method, characterized in that: include: Acquire historical trajectory information of the target vehicle to be predicted and surrounding vehicles, and determine first trajectory characteristics and kinematic characteristics of the target vehicle to be predicted and second trajectory characteristics of the surrounding vehicles based on the historical trajectory information; Inputting the first trajectory feature and the kinematic feature into a time attention module, and determining the time attention weight of the first trajectory feature of the target vehicle to be predicted at a historical moment and the first trajectory feature at a current moment through the time attention module; Inputting the first trajectory feature, the kinematic feature and the second trajectory feature into a spatial attention module, and determining the spatial attention weights of the surrounding vehicles and the target vehicle to be predicted by the spatial attention module; The spatiotemporal feature information of the target vehicle to be predicted is determined according to the temporal attention weight and the spatial attention weight, and the trajectory prediction result of the target vehicle to be predicted is outputted according to the spatiotemporal feature information through a decoder module.

2. The method according to claim 1, characterized in that The acquiring historical trajectory information of the target vehicle to be predicted and surrounding vehicles, and determining the first trajectory feature and the kinematic feature of the target vehicle to be predicted and the second trajectory feature of the surrounding vehicles based on the historical trajectory information respectively, includes: The historical trajectory information of the target vehicle to be predicted and the surrounding vehicles are respectively embedded by a nonlinear activation function, and the embedded historical trajectory information is encoded by a long short-term memory network to obtain a first trajectory feature of the target vehicle to be predicted and a second trajectory feature of the surrounding vehicles; The kinematic characteristics of the target vehicle to be predicted are determined based on the historical trajectory information of the target vehicle to be predicted through a fully connected layer network.

3. The method according to claim 1, characterized in that The determining of the time attention weights of the historical moment trajectory and the current moment trajectory of the target vehicle to be predicted by the time attention module includes: The time correlation between the first trajectory feature of the target vehicle to be predicted at the current moment and the first trajectory feature of each historical moment is calculated respectively by using a similarity function; The time attention weights of the first trajectory features of each historical moment and the first trajectory features of the current moment are calculated according to the time correlation through an activation function and a normalized exponential function, wherein the time attention module includes a similarity function, an activation function and a normalized exponential function.

4. The method according to claim 1, characterized in that: The determining, by the spatial attention module, the spatial attention weights of the surrounding vehicles and the target vehicle to be predicted includes: Calculating the spatial correlation between the first trajectory feature and the second trajectory feature by using a cosine distance metric function; The spatial attention weights of the surrounding vehicles and the target vehicle to be predicted are calculated according to the spatial correlation through a normalization function, wherein the spatial attention module includes a cosine distance measurement function and a normalization function.

5. The method according to claim 1, characterized in that The step of determining the spatiotemporal feature information of the target vehicle to be predicted according to the temporal attention weight and the spatial attention weight, and outputting the trajectory prediction result of the target vehicle to be predicted according to the spatiotemporal feature information through a decoder module includes: Determine the prediction coordinate points based on the preconfigured prediction time domain and sampling frequency; For each of the predicted coordinate points, a context vector containing temporal attention information is calculated according to the temporal attention weight, and a context vector containing spatial attention information is calculated according to the spatial attention weight; Determine the spatiotemporal feature information of the target vehicle to be predicted corresponding to the predicted coordinate point through a fully connected layer network according to the context vector containing the temporal attention information and the context vector containing the spatial attention information of each predicted coordinate point; The trajectory prediction result of the target vehicle to be predicted is outputted through the decoder module according to the spatiotemporal feature information of each prediction coordinate point.

6. The method according to claim 5, characterized in that The step of calculating a context vector containing temporal attention information according to the temporal attention weight, and calculating a context vector containing spatial attention information according to the spatial attention weight, comprises: Calculating a context vector containing temporal attention information according to the first trajectory features of each historical moment and the temporal attention weight by the temporal attention module; The spatial attention module calculates a context vector containing spatial attention information according to the second trajectory features of the surrounding vehicles and the spatial attention weights.

7. The method according to claim 1, characterized in that The decoder module outputs the trajectory prediction result of the target vehicle to be predicted according to the spatiotemporal feature information of each prediction coordinate point, including: The decoder module maps the spatiotemporal feature information of the target vehicle to be predicted at each prediction coordinate point to the driving behavior space to obtain the driving behavior prediction result of each prediction coordinate point; Determine the driving trajectory of the target vehicle to be predicted according to the driving behavior prediction results of each prediction coordinate point; Determine spatial attention weight distribution information according to the spatial attention weight of each predicted coordinate point, determine temporal attention weight distribution information according to the temporal attention weight of each predicted coordinate point, and output the driving trajectory, driving behavior prediction result, spatial attention weight distribution information and temporal attention weight distribution information; The decoder module is constructed based on the following driving behaviors: longitudinal acceleration, constant speed, deceleration, lateral left lane change, lane keeping and right lane change.

8. A vehicle driving trajectory prediction device, characterized in that: include: A historical trajectory information acquisition module is used to acquire historical trajectory information of the target vehicle to be predicted and surrounding vehicles, and determine the first trajectory characteristics and kinematic characteristics of the target vehicle to be predicted and the second trajectory characteristics of the surrounding vehicles based on the historical trajectory information; a time attention weight determination module, configured to input the first trajectory feature and the kinematic feature into a time attention module, and determine the time attention weight of the first trajectory feature at a historical moment and the first trajectory feature at a current moment of the target vehicle to be predicted through the time attention module; a spatial attention weight determination module, configured to input the first trajectory feature, the kinematic feature, and the second trajectory feature into a spatial attention module, and determine the spatial attention weights of the surrounding vehicles and the target vehicle to be predicted through the spatial attention module; The trajectory prediction result output module is used to determine the spatiotemporal feature information of the target vehicle to be predicted according to the temporal attention weight and the spatial attention weight, and output the trajectory prediction result of the target vehicle to be predicted according to the spatiotemporal feature information through a decoder module.

9. A car, characterized in that: The automobile comprises: at least one processor; and a memory communicatively connected to the at least one processor; wherein, The memory stores a computer program executable by the at least one processor, and the computer program is executed by the at least one processor so that the at least one processor can execute a vehicle driving trajectory prediction method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer instructions, and the computer instructions are used to enable a processor to implement a vehicle driving trajectory prediction method according to any one of claims 1 to 7 when executed.

Images