A method, device, electronic device and storage medium for predicting vehicle intentions

Abstract

The present application relates to a method, apparatus, electronic device and storage medium for predicting vehicle intentions, and relates to the field of computer technology. The method includes: obtaining an intention grid corresponding to a target vehicle, and then predicting the vehicle intention of the target vehicle based on the intention grid corresponding to the target vehicle. The method, apparatus, electronic device and storage medium for predicting vehicle intentions provided by the present application can realize the prediction of the vehicle intention of the target vehicle.

Description

Technical Field

[0001] The present application relates to the field of computer technology, and in particular, to a method, apparatus, electronic device, and storage medium for predicting vehicle intentions. Background Art

[0002] Trajectory prediction tasks are of great significance for driverless driving, and this will still be a key issue to be solved in the driverless industry. In different scenarios, a vehicle may have different possible driving intentions. For example, in a highway scenario, the feasible driving intentions are usually divided into three types: going straight, changing lanes to the left, and changing lanes to the right; while at an intersection, in addition to the three types listed above, the feasible driving intentions also include turning left, turning right, and going straight through the intersection.

[0003] Considering that driving intentions have a great impact on vehicle safety and vehicle control, etc., potential collision risks can be reduced, such as vehicle trajectory prediction based on driving intentions, vehicle deviation warning based on driving intentions, and vehicle collision warning based on vehicle intentions, etc. Therefore, how to determine the intention of a vehicle has become a key issue. Summary of the Invention

[0004] The purpose of the present application is to provide a method, apparatus, electronic device, and storage medium for predicting vehicle intentions, which can be used to solve the above technical problems.

[0005] The above invention purpose of the present application is achieved through the following technical solutions:

[0006] In a first aspect, a method for predicting vehicle intentions is provided, including:

[0007] Obtain an intention grid corresponding to a target vehicle;

[0008] Based on the intention grid corresponding to the target vehicle, perform vehicle intention prediction on the target vehicle;

[0009] Among them, the generation method of the intention grid includes:

[0010] Obtain a trajectory data set and each scenario map, and obtain the historical vehicle trajectories corresponding to each scenario from the trajectory data set. The historical vehicle trajectories include the historical trajectories corresponding to multiple vehicles respectively, and the historical trajectory corresponding to any vehicle is composed of multiple historical trajectory points;

[0011] Screen the historical trajectory points in the preset area of each scenario map from the historical vehicle trajectories corresponding to each scenario respectively, and determine the intention information corresponding to each vehicle passing through each historical trajectory point;

[0012] Based on the intention information corresponding to each vehicle passing through each historical trajectory point, determine the intention information corresponding to each historical trajectory point;

[0013] Generate an intention grid corresponding to each scenario based on the intention information corresponding to each historical trajectory point and the position information corresponding to each historical trajectory point respectively, where the intention grid contains the position information corresponding to each historical trajectory point and the intention information corresponding to each of them.

[0014] In a possible implementation, before predicting the vehicle intention of the target vehicle based on the intention grid corresponding to the target vehicle, it further includes:

[0015] Obtain the historical trajectory information of the target vehicle;

[0016] Among them, predicting the vehicle intention of the target vehicle based on the intention grid corresponding to the target vehicle includes:

[0017] Predict the vehicle intention of the target vehicle based on the historical trajectory information of the target vehicle and the intention grid corresponding to the target vehicle.

[0018] In another possible implementation, predicting the vehicle intention of the target vehicle based on the historical trajectory information of the target vehicle and the intention grid includes:

[0019] Predict the vehicle intention at the current moment based on the historical trajectory information of the target vehicle and the intention grid and through a recurrent neural network and a multi-layer perceptron.

[0020] In another possible implementation, the historical trajectory information of the target vehicle consists of multiple historical trajectory points, the intention grid contains multiple grids, and any grid contains multiple grid points;

[0021] Predicting the vehicle intention at the current moment based on the historical trajectory information of the target vehicle and the intention grid and through a recurrent neural network and a multi-layer perceptron includes:

[0022] Obtain the position coordinates corresponding to the target vehicle at the current moment;

[0023] Predict the state information corresponding to the target vehicle at the current moment by passing the position coordinates corresponding to the target vehicle at the current moment and the position coordinates corresponding to each historical trajectory point respectively through a recurrent neural network;

[0024] Based on the intention grid, determine the position information and intention information corresponding to each grid point in the grid where the target vehicle is currently located;

[0025] Predict the vehicle intention at the current moment by passing the position information and intention information corresponding to each grid point respectively and the state information corresponding to the target vehicle at the current moment through a multi-layer perceptron.

[0026] In another possible implementation, before predicting the position coordinates corresponding to the target vehicle at the current moment and the position coordinates corresponding to each historical trajectory point through a recurrent neural network to obtain the state information corresponding to the target vehicle at the current moment, it further includes:

[0027] Converting the position coordinates corresponding to each historical trajectory point from global coordinates to local coordinates; and,

[0028] Converting the position coordinates corresponding to the target vehicle at the current moment from global coordinates to local coordinates;

[0029] Wherein, predicting the position coordinates corresponding to the target vehicle at the current moment and the position coordinates corresponding to each historical trajectory point through a recurrent neural network to obtain the state information corresponding to the target vehicle at the current moment includes:

[0030] Predicting the local coordinates corresponding to the target vehicle at the current moment and the local coordinates corresponding to each historical trajectory point through a recurrent neural network to obtain the state information corresponding to the target vehicle at the current moment.

[0031] In another possible implementation, predicting the vehicle intention at the current moment by using a multi-layer perceptron for the position information, intention information, and state information corresponding to each historical trajectory point in the grid where the target vehicle is currently located and the state information corresponding to the target vehicle at each historical moment includes:

[0032] Encoding the state information corresponding to the target vehicle at the current moment to obtain the encoded state information;

[0033] Encoding the position information and intention information corresponding to each grid point in the grid where the target vehicle is currently located to obtain the encoded grid information;

[0034] Connecting the encoded state information and the encoded grid information to obtain the connected encoded information;

[0035] Predicting the vehicle intention at the current moment based on the connected encoded information through the multi-layer perceptron.

[0036] In another possible implementation, the method further includes:

[0037] Obtaining the position coordinates of the trajectory point of the target vehicle at the current moment;

[0038] Obtaining the historical trajectory features of the target vehicle;

[0039] Obtain the interaction features of the target vehicle at the current moment, where the interaction features of the target vehicle at the current moment are used to characterize the interaction relationship between the target vehicle and other vehicles at the current moment;

[0040] Predict the vehicle trajectory information of the target vehicle based on the position coordinates of the trajectory points of the target vehicle at the current moment, the historical trajectory features of the target vehicle, and the interaction features at the current moment.

[0041] In another possible implementation, before obtaining the interaction features of the target vehicle at the current moment, it further includes:

[0042] Obtain the position information corresponding to each vehicle at the current moment;

[0043] Select the position information of the preset vehicle from the position information corresponding to each vehicle at the current moment;

[0044] Determine the lane information where each selected vehicle is located at the current moment;

[0045] Generate the interaction features of the target vehicle at the current moment based on the position information of the preset vehicle and the lane information where each selected vehicle is located at the current moment.

[0046] In another possible implementation, the generating the interaction features of the target vehicle at the current moment based on the position information of the preset vehicle and the lane information where each selected vehicle is located at the current moment includes:

[0047] Connect the position information of each vehicle at the current moment with the lane information where each vehicle is located at the current moment;

[0048] Perform encoding processing on the connected vehicle information to obtain the interaction features of the target vehicle at the current moment.

[0049] In another possible implementation, before obtaining the historical trajectory features of the target vehicle, it further includes:

[0050] Obtain the historical trajectory information of the target vehicle, where the historical trajectory information is composed of multiple historical trajectory points;

[0051] Perform encoding processing on the position coordinates corresponding to each historical trajectory point to obtain the historical trajectory features of the target vehicle.

[0052] In another possible implementation, the predicting the vehicle trajectory information of the target vehicle based on the trajectory points of the target vehicle at the current moment, the historical trajectory features of the target vehicle, and the interaction features at the current moment includes:

[0053] Perform trajectory prediction processing on the historical trajectory features of the target vehicle and the trajectory points of the target vehicle at the current moment to obtain the trajectory prediction result at the next moment;

[0054] For each target moment after the next moment, perform the following steps in sequence according to the time sequence until the obtained trajectory prediction result meets the preset conditions:

[0055] Predict the vehicle intention at the previous moment of the target moment based on the trajectory prediction result at the previous moment of the target moment. For the first target moment, the trajectory prediction result at the previous moment of the target moment is the trajectory prediction result at the next moment;

[0056] Perform trajectory prediction processing based on the trajectory prediction result at the previous moment of the target moment, the interaction features of the target vehicle at the current moment, and the vehicle intention at the previous moment of the target moment to obtain the trajectory prediction result at the target moment.

[0057] In another possible implementation, the performing trajectory prediction processing on the historical trajectory features of the target vehicle and the trajectory points of the target vehicle at the current moment to obtain the trajectory prediction result at the next moment includes:

[0058] Perform encoding processing according to the historical trajectory features of the target vehicle and the trajectory points of the target vehicle at the current moment to obtain the hidden layer variable at the current moment;

[0059] Perform encoding processing based on the hidden layer variable at the current moment to obtain the trajectory prediction result at the next moment.

[0060] In another possible implementation, the performing trajectory prediction processing based on the trajectory prediction result at the previous moment of the target moment, the interaction features of the target vehicle at the current moment, and the vehicle intention at the previous moment of the target moment to obtain the trajectory prediction result at the target moment includes:

[0061] Perform encoding through an attention mechanism based on the trajectory prediction result at the previous moment of the target moment, the interaction features of the target vehicle at the current moment, and the vehicle intention at the previous moment of the target moment to obtain the encoding result of the prediction point at the previous moment of the target moment;

[0062] Perform trajectory prediction processing based on the encoding result of the prediction point to obtain the trajectory prediction result at the target moment.

[0063] In a second aspect, there is provided a device for predicting vehicle intention, including:

[0064] A first acquisition module for acquiring an intention grid corresponding to the target vehicle;

[0065] A vehicle intention prediction module for predicting the intention of a target vehicle based on the intention grid corresponding to the target vehicle;

[0066] Among them, when the first generation module generates the intention grid, it is specifically used for:

[0067] Obtain a trajectory data set and each scenario map, and obtain the historical vehicle trajectories corresponding to each scenario from the trajectory data set. The historical vehicle trajectories include the historical trajectories corresponding to multiple vehicles respectively, and the historical trajectory corresponding to any vehicle is composed of multiple historical trajectory points;

[0068] Filter the historical trajectory points in the preset area of each scenario map from the historical vehicle trajectories corresponding to each scenario respectively, and determine the intention information corresponding to each vehicle passing through each historical trajectory point;

[0069] Based on the intention information corresponding to each vehicle passing through each historical trajectory point, determine the intention information corresponding to each historical trajectory point;

[0070] Based on the intention information corresponding to each historical trajectory point and the position information corresponding to each historical trajectory point respectively, generate the intention grids corresponding to each scenario. The intention grid contains the position information corresponding to each historical trajectory point and the intention information corresponding to each of them.

[0071] In a possible implementation manner, the device further includes: a second acquisition module, among which,

[0072] The second acquisition module is used to acquire the historical trajectory information of the target vehicle;

[0073] Among them, when the vehicle intention prediction module predicts the intention of the target vehicle based on the intention grid corresponding to the target vehicle, it is specifically used for:

[0074] Predict the intention of the vehicle at the current moment based on the historical trajectory information of the target vehicle and the intention grid corresponding to the target vehicle.

[0075] In another possible implementation manner, when the vehicle intention prediction module predicts the intention of the target vehicle based on the historical trajectory information and the intention grid of the target vehicle, it is specifically used for:

[0076] Predict the intention of the vehicle at the current moment based on the historical trajectory information of the target vehicle and the intention grid through a recurrent neural network and a multi-layer perceptron.

[0077] In another possible implementation manner, the historical trajectory information of the target vehicle is composed of multiple historical trajectory points, the intention grid contains multiple grids, and any grid contains multiple grid points;

[0078] When the vehicle intention prediction module predicts the vehicle intention at the current moment based on the historical trajectory information of the target vehicle and the intention grid, and through a recurrent neural network and a multi-layer perceptron, it is specifically used for:

[0079] Obtain the position coordinates corresponding to the target vehicle at the current moment;

[0080] Predict the position coordinates corresponding to the target vehicle at the current moment and the position coordinates corresponding to each historical trajectory point through a recurrent neural network to obtain the state information corresponding to the target vehicle at each historical moment;

[0081] Based on the intention grid, determine the position information and intention information corresponding to each grid point in the grid where the target vehicle is currently located;

[0082] Predict the vehicle intention at the current moment through a multi-layer perceptron using the position information and intention information corresponding to each grid point and the state information corresponding to the target vehicle at the current moment.

[0083] In another possible implementation, the device further includes: a conversion module, where,

[0084] The conversion module is used to convert the position coordinates corresponding to each historical trajectory point from global coordinates to local coordinates; and,

[0085] Convert the position coordinates corresponding to the target vehicle at the current moment from global coordinates to local coordinates;

[0086] Wherein, when the vehicle intention prediction module predicts the state information corresponding to the target vehicle at the current moment by predicting the position coordinates corresponding to each historical trajectory point through a recurrent neural network, it is specifically used for:

[0087] Predict the state information corresponding to the target vehicle at the current moment by predicting the local coordinates corresponding to the target vehicle at the current moment and the local coordinates corresponding to each historical trajectory point through a recurrent neural network.

[0088] In another possible implementation, when the vehicle intention prediction module predicts the vehicle intention at the current moment through a multi-layer perceptron using the position information and intention information corresponding to each historical trajectory point in the grid where the target vehicle is currently located and the state information corresponding to the target vehicle at each historical moment, it is specifically used for:

[0089] Perform encoding processing on the state information corresponding to the target vehicle at the current moment to obtain the encoded state information;

[0090] Encode the position information and intention information corresponding to each grid point in the grid where the target vehicle is currently located to obtain the encoded grid information;

[0091] Connect the encoded state information and the encoded grid information to obtain the connected encoded information;

[0092] Predict the vehicle intention at the current moment based on the connected encoded information and through the multi-layer perceptron.

[0093] In another possible implementation, the device further includes: a third acquisition module, a fourth acquisition module, a fifth acquisition module, and a vehicle trajectory prediction module, where,

[0094] The third acquisition module is used to acquire the position coordinates of the trajectory points of the target vehicle at the current moment;

[0095] The fourth acquisition module is used to acquire the historical trajectory features of the target vehicle;

[0096] The fifth acquisition module is used to acquire the interaction features of the target vehicle at the current moment, and the interaction features of the target vehicle at the current moment are used to characterize the interaction relationship between the target vehicle and other vehicles at the current moment;

[0097] The vehicle trajectory prediction module is used to predict the vehicle trajectory information of the target vehicle based on the position coordinates of the trajectory points of the target vehicle at the current moment, the historical trajectory features of the target vehicle, and the interaction features at the current moment.

[0098] In another possible implementation, the device further includes: a sixth acquisition module, a selection module, a determination module, and a generation module, where,

[0099] The sixth acquisition module is used to acquire the position information corresponding to all vehicles at the current moment;

[0100] The selection module is used to select the position information of the preset vehicle from the position information corresponding to all vehicles at the current moment;

[0101] The determination module is used to determine the lane information where each selected vehicle is located at the current moment;

[0102] The generation module is used to generate the interaction features of the target vehicle at the current moment based on the position information of the preset vehicle and the lane information where each selected vehicle is located at the current moment.

[0103] In another possible implementation manner, when generating the interaction feature of the target vehicle at the current moment based on the position information of the preset vehicle and the lane information of each selected vehicle at the current moment respectively, the generating module is specifically configured to:

[0104] Connect the position information of each vehicle at the current moment with the lane information of each vehicle at the current moment;

[0105] Encode the connected vehicle information to obtain the interaction feature of the target vehicle at the current moment.

[0106] In another possible implementation manner, the device further includes: a seventh acquisition module and an encoding processing module, where,

[0107] The seventh acquisition module is configured to acquire the historical trajectory information of the target vehicle, and the historical trajectory information is composed of a plurality of historical trajectory points;

[0108] The encoding processing module is configured to encode the position coordinates corresponding to each historical trajectory point to obtain the historical trajectory feature of the target vehicle.

[0109] In another possible implementation manner, when predicting the vehicle trajectory information of the target vehicle based on the trajectory points of the target vehicle at the current moment, the historical trajectory feature of the target vehicle, and the interaction feature at the current moment, the vehicle trajectory prediction module is specifically configured to:

[0110] Perform trajectory prediction processing on the historical trajectory feature of the target vehicle and the trajectory points of the target vehicle at the current moment to obtain the trajectory prediction result at the next moment;

[0111] For each target moment after the next moment, perform the following steps in sequence according to the time sequence until the obtained trajectory prediction result meets the preset conditions:

[0112] Predict the vehicle intention at the previous moment of the target moment based on the trajectory prediction result at the previous moment of the target moment. For the first target moment, the trajectory prediction result at the previous moment of the target moment is the trajectory prediction result at the next moment;

[0113] Perform trajectory prediction processing based on the trajectory prediction result at the previous moment of the target moment, the interaction feature of the target vehicle at the current moment, and the vehicle intention at the previous moment of the target moment to obtain the trajectory prediction result at the target moment.

[0114] In another possible implementation, when the vehicle trajectory prediction module performs trajectory prediction processing on the historical trajectory features of the target vehicle and the trajectory points of the target vehicle at the current moment to obtain the trajectory prediction result at the next moment, it is specifically configured to:

[0115] Perform encoding processing according to the historical trajectory features of the target vehicle and the trajectory points of the target vehicle at the current moment to obtain the hidden layer variable at the current moment;

[0116] Perform encoding processing based on the hidden layer variable at the current moment to obtain the trajectory prediction result at the next moment.

[0117] In another possible implementation, when the vehicle trajectory prediction module performs trajectory prediction processing based on the trajectory prediction result at the previous moment of the target moment, the interaction features of the target vehicle at the current moment, and the vehicle intention at the previous moment of the target moment to obtain the trajectory prediction result at the target moment, it is specifically configured to:

[0118] Perform encoding through an attention mechanism based on the trajectory prediction result at the previous moment of the target moment, the interaction features of the target vehicle at the current moment, and the vehicle intention at the previous moment of the target moment to obtain the encoding result of the prediction point at the previous moment of the target moment;

[0119] Perform trajectory prediction processing based on the encoding result of the prediction point to obtain the trajectory prediction result at the target moment.

[0120] In a third aspect, an electronic device is provided, and the electronic device includes:

[0121] One or more processors;

[0122] A memory;

[0123] One or more applications, where one or more applications are stored in the memory and are configured to be executed by one or more processors, and one or more programs are configured to: execute the operations corresponding to the method for predicting vehicle intention shown in any possible implementation of the first aspect.

[0124] In a fourth aspect, a computer-readable storage medium is provided, and the storage medium stores at least one instruction, at least one segment of program, a code set or an instruction set, and at least one instruction, at least one segment of program, the code set or the instruction set is loaded and executed by a processor to implement the method for predicting vehicle intention shown in any possible implementation of the first aspect.

[0125] In summary, the present application includes at least one of the following beneficial technical effects:

[0126] The present application provides a method, apparatus, electronic device, and storage medium for vehicle intention prediction. Compared with the related art, in the present application, by obtaining the historical vehicle trajectories corresponding to each scenario map and screening out the historical trajectory points within a preset area in each scenario map, and then further determining the intention information corresponding to each vehicle passing through each trajectory point, the intention information corresponding to each trajectory point can be statistically obtained. Thus, according to the position information and intention information of each trajectory point, corresponding intention grids for each scenario are generated, so that when predicting the intention of a target vehicle, the intention grid corresponding to the target vehicle can be used to achieve vehicle intention prediction. BRIEF DESCRIPTION OF THE DRAWINGS

[0127] Figure 1a is a schematic flowchart of a method for vehicle intention prediction provided by an embodiment of the present application;

[0128] Figure 1b is a schematic flowchart of a method for generating an intention grid provided by an embodiment of the present application;

[0129] Figure 2 is a schematic diagram of intention statistics in an intersection scenario provided by an embodiment of the present application;

[0130] Figure 3 is a schematic flowchart of the intention prediction of a target vehicle provided by an embodiment of the present application;

[0131] Figure 4 is a schematic diagram of an intention grid sampled from around a target vehicle provided by an embodiment of the present application;

[0132] Figure 5 is another schematic diagram of the intention prediction of a target vehicle provided by an embodiment of the present application;

[0133] Figure 6 is a schematic diagram of obtaining the interaction information of a target vehicle provided by an embodiment of the present application;

[0134] Figure 7 is a schematic diagram of vehicle trajectory generation provided by an embodiment of the present application;

[0135] Figure 8 is a schematic structural diagram of a device for vehicle intention prediction provided by an embodiment of the present application;

[0136] Figure 9 is a schematic structural diagram of an electronic device provided by an embodiment of the present application. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0137] The following further describes the present application in detail with reference to the accompanying drawings.

[0138] This specific embodiment is only an interpretation of the present application and does not limit the present application. After reading this specification, those skilled in the art can make modifications to this embodiment without creative contributions as needed, but as long as it is within the scope of the claims of the present application, it is protected by the patent law.

[0139] An embodiment of the present application provides a method for predicting vehicle intentions. The surrounding environment of the vehicle's location is used as one of the inputs for intention prediction. An intention grid is generated based on a large number of vehicle historical trajectories, and intention prediction is performed according to the intention grid. After obtaining the intention output, the intention prediction result can be used for vehicle trajectory prediction and vehicle deviation warning, etc. Specifically, machine learning methods can be used for trajectory prediction and vehicle deviation warning, etc., so that the prediction method can be applied to different environments and improve the accuracy of trajectory prediction and vehicle deviation, etc. In the embodiment of the present application, intention prediction can be taken as an example for trajectory prediction.

[0140] Specifically, an embodiment of the present application provides a vehicle intention method and a trajectory prediction method. Using the intention information when the vehicle passes through an intersection, sampling is performed on the map to generate a grid with intention labels. Based on the intention grid and its own vehicle trajectory, a recurrent neural network is used for intention prediction. After obtaining the intention prediction result, combined with its own vehicle trajectory and the trajectories of surrounding vehicles, an attention mechanism and a recurrent neural network are used to complete trajectory prediction.

[0141] To make the objectives, technical solutions, and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are some, but not all, of the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without making creative efforts belong to the scope of protection of the present application.

[0142] In addition, the term "and / or" in this article is only a description of the association relationship between associated objects, indicating that there can be three relationships. For example, A and / or B can represent: A exists alone, A and B exist simultaneously, and B exists alone. In addition, the character " / " in this article, unless otherwise specified, generally represents an "or" relationship between the associated objects before and after.

[0143] The embodiments of the present application will be further described in detail below with reference to the accompanying drawings of the specification.

[0144] An embodiment of the present application provides a method, device, electronic device, and storage medium for vehicle intention prediction. The method for vehicle intention prediction can be executed by an electronic device, which can be a server or a terminal device. Among them, the server can be an independent physical server, a server cluster or a distributed system composed of multiple physical servers, or a cloud server providing cloud computing services. The terminal device can be a smart phone, a tablet computer, a notebook computer, a desktop computer, etc., but is not limited thereto. The terminal device and the server can be directly or indirectly connected through wired or wireless communication methods, and the embodiments of the present application do not limit this.

[0145] It should be noted that the electronic device for executing the method of vehicle prediction may further include: in-vehicle devices installed on various intelligent vehicles, intelligent robots, etc.

[0146] Furthermore, as Figure 1a shown, the method may include:

[0147] Step S101, obtain an intention grid corresponding to the target vehicle.

[0148] For the embodiments of the present application, the intention grid is used to characterize each position in the grid and the vehicle intentions respectively corresponding to each position. In the embodiments of the present application, intention grids corresponding to each scenario can be generated in advance. When performing vehicle intention prediction on a vehicle, determine the current scenario where the vehicle is located, and then obtain the intention grid corresponding to the current scenario for vehicle intention prediction; or directly obtain the intention grids corresponding to each scenario to perform vehicle intention prediction on the target vehicle. For example, the scenarios where the vehicle is located may include: intersection scenarios, T-junction scenarios, highway scenarios, etc.

[0149] Specifically, when it is necessary to perform vehicle intention prediction on the target vehicle, obtain the current environment image based on the camera device installed on the target vehicle, and analyze the current environment image to determine the current scenario where the target vehicle is located; or when it is necessary to perform vehicle intention prediction on the target vehicle, determine the current position information corresponding to the target vehicle, and based on the current position information corresponding to the target vehicle, determine the current scenario where the target vehicle is located through a map, etc.; or obtain the scenario information input by the user and determine the scenario information input by the user as the current scenario where the target vehicle is located.

[0150] Before obtaining the intention grid corresponding to the target vehicle, it is also necessary to generate intention grids corresponding to each scenario. Among them, the generation method of the intention grid can specifically include: Step Sa, Step Sb, Step Sc, and Step Sd, as Figure 1b shown, including:

[0151] Step Sa: Obtain the trajectory dataset and each scene map, and obtain the historical vehicle trajectories corresponding to each scene from the trajectory dataset.

[0152] The trajectory dataset can be the NGSIM dataset or other datasets, where the trajectory dataset is a dataset including vehicle information and vehicle trajectory information.

[0153] Among them, the vehicle information includes the type, length, and width of the vehicle. The vehicle trajectory information includes lateral displacement, longitudinal displacement, vehicle speed, acceleration, distance from the vehicle in front, lane where the vehicle is located, and vehicle intention. The dataset also contains maps of highways and urban roads.

[0154] Furthermore, the historical vehicle trajectories include the historical trajectories corresponding to multiple vehicles respectively, and the historical trajectory corresponding to any vehicle is composed of multiple historical trajectory points.

[0155] Step Sb: Screen the historical trajectory points within the preset area in each scene map from the historical vehicle trajectories corresponding to each scene respectively, and determine the intention information corresponding to each vehicle passing through each historical trajectory point.

[0156] After obtaining the historical vehicle trajectories corresponding to each scene respectively, determine the historical trajectory points within the preset area according to the historical trajectory information corresponding to each scene, and determine the intention information corresponding to each vehicle passing through each historical trajectory point within the preset area. As Figure 2 shown, count the vehicle trajectories within the boxed area at the intersection. The boxed area consists of two rectangles. One rectangle is the central area of the intersection, which is a multi-lane intersection, and a part is the parking and waiting area. Here, consider the vehicles passing from south to north (from bottom to top). Screen the trajectory dataset. First, screen out all the trajectory points within the boxed range, that is, the trajectories with coordinates within the boxed range, and then determine the intention information corresponding to the vehicles passing through each trajectory point within the boxed area.

[0157] Furthermore, in order to further reduce the data volume and improve the accuracy of intention prediction, screen the vehicles based on the differences in scenes and boxed areas to determine the intentions of the screened vehicles. As Figure 2 shown, next, screen the vehicles traveling northward. The Direction label is 2 - Northbound (NB). Next, count the number of vehicles with different intentions at each coordinate point respectively, in order to Figure 2Taking the selected area as an example, the lower left coordinate of the upper matrix is (2230492, 1375670), and the upper right coordinate is (2230551, 1375746), with a total of 4484 points. The lower left coordinate of the lower matrix is (2230511, 1375605), and the upper right coordinate is (2230544, 1375667), with a total of 2046 points. The entire area has a total of 6530 points. For each point, the intention of the vehicles passing through this point at all times is counted. In the dataset, 1, 2, and 3 represent that the vehicle goes straight, turns left, and turns right respectively. A piecewise function is designed to convert the intention label to 0, -1, and +1 to represent going straight, turning left, and turning right. Among them,

[0158] ;

[0159] Among them, represents the intention of this point.

[0160] Step Sc: Based on the intention information corresponding to each vehicle passing through each historical trajectory point, determine the intention information corresponding to each historical trajectory point.

[0161] After obtaining the intention information corresponding to each vehicle passing through each historical trajectory point, calculate the average value of the intention label values of all times and vehicles at each historical trajectory point, as shown in the following formula, to obtain a value between -1 and 1, which represents the vehicle intention. The closer it is to 0, the greater the proportion of vehicles going straight through this point; the closer it is to -1, the greater the proportion of vehicles turning left; the closer it is to 1, the greater the proportion of vehicles turning right.

[0162]

[0163] Among them, is the intention label of the vehicle at all times at the coordinate , with a total of N labels, is the intention value at the coordinate .

[0164] Step Sd: Based on the intention information corresponding to each historical trajectory point and the position information corresponding to each historical trajectory point, generate the intention grid corresponding to each scenario.

[0165] Among them, the intention grid contains the position information corresponding to each historical trajectory point and its respective corresponding intention information.

[0166] After obtaining the intention value and the corresponding position coordinates corresponding to each historical trajectory point through the above embodiments, concatenate the intention value and its coordinates to form a vector d, where,

[0167]

[0168] Among them, d represents a grid point, which includes the abscissa and ordinate of the grid point and the intention value. Where x and y are the abscissa and ordinate, and m is a value between -1 and 1, representing the vehicle intention;

[0169] ;

[0170] The set of all grid points in the scene forms the intention grid D, which has 6,530 groups of values, representing the coordinates and intention values of each point in the selected area.

[0171] Step S102: Based on the intention grid corresponding to the target vehicle, perform vehicle intention prediction on the target vehicle.

[0172] After obtaining the intention grid corresponding to the target vehicle through the above embodiments, perform vehicle intention prediction on the target vehicle to obtain the vehicle intention corresponding to the target vehicle. For example, in the scenario of an intersection, the vehicle intention corresponding to the target vehicle may include going straight, turning left, or turning right; in the scenario of a general normal multi-lane road, the intention label corresponding to the target vehicle may include: parallel or going straight; in the scenario of the entrance and exit of a highway, the intention label corresponding to the target vehicle may include: going straight or entering / leaving the ramp.

[0173] The embodiment of the present application provides a method for vehicle intention prediction. Compared with the related art, in the embodiment of the present application, by obtaining the historical vehicle trajectories corresponding to each scene map respectively, and screening out the historical trajectory points belonging to the preset area in each scene map, and then further determining the intention information corresponding to each vehicle passing through each trajectory point, the intention information corresponding to each trajectory point can be counted, so that according to the position information and intention information of each trajectory point, the corresponding intention grids for each scene are generated, so that when performing intention prediction on the target vehicle, the vehicle intention prediction can be realized based on the intention grid corresponding to the target vehicle.

[0174] Further, in order to improve the accuracy of vehicle intention prediction for the target vehicle, before performing vehicle intention prediction on the target vehicle based on the intention grid corresponding to the target vehicle in step S102, it may further include: obtaining the historical trajectory information of the target vehicle. In the embodiment of the present application, the historical trajectory information of the target vehicle may include: the coordinate information corresponding to each historical trajectory point.

[0175] Further, on the basis of obtaining the historical trajectory information of the target vehicle, when performing vehicle intention prediction on the target vehicle based on the intention grid corresponding to the target vehicle in step S102, it may specifically include: performing vehicle intention prediction on the target vehicle based on the historical trajectory information of the target vehicle and the intention grid corresponding to the target vehicle.

[0176] Specifically, vehicle intention prediction is performed on the target vehicle based on the historical trajectory information of the target vehicle and the intention grid corresponding to the target vehicle, which may specifically include: encoding the historical trajectory information of the target vehicle to obtain the historical trajectory features of the target vehicle, and performing vehicle intention prediction on the target vehicle based on the historical trajectory features of the target vehicle and the intention grid corresponding to the target vehicle.

[0177] Further, vehicle intention prediction is performed on the target vehicle based on the historical trajectory information of the target vehicle and the intention grid corresponding to the target vehicle, which may specifically include: predicting the vehicle intention at the current moment based on the historical trajectory information of the target vehicle, the intention grid, a recurrent neural network, and a multi-layer perceptron.

[0178] Specifically, the historical trajectory information of the target vehicle consists of multiple historical trajectory points; predicting the vehicle intention at the current moment based on the historical trajectory information of the target vehicle, the intention grid, a recurrent neural network, and a multi-layer perceptron may specifically include: obtaining the position coordinates corresponding to the target vehicle at the current moment; predicting the state information corresponding to the target vehicle at the current moment through the recurrent neural network using the position coordinates corresponding to the target vehicle at the current moment and the position coordinates corresponding to each historical trajectory point respectively; determining the position information and intention information corresponding to each grid point in the grid where the target vehicle is currently located based on the intention grid; and predicting the vehicle intention at the current moment through the multi-layer perceptron using the position information and intention information corresponding to each grid point in the grid where the target vehicle is currently located and the state information corresponding to the target vehicle at the current moment.

[0179] As Figure 3 shown, X can be used to represent the position information corresponding to the historical trajectory points of the target vehicle and the position coordinates corresponding to the target vehicle at the current moment, where the position coordinates corresponding to the target vehicle at the current moment are represented by and the position coordinates corresponding to the target vehicle at the t-1 moment are represented by i.e., used to represent the position information corresponding to the historical trajectory points of the target vehicle, where x t , y t are respectively the abscissa and ordinate of the target vehicle at the t-th moment (current moment), and x t-1 , y t-1 are respectively the abscissa and ordinate of the target vehicle at the t-1 moment. Then, the trajectory points are input into the RNN recurrent neural network, and the RNN is used to predict the target point at this moment. The specific prediction method is shown in the following formula:

[0180] ;

[0181] where the specific implementation method of the RNN is

[0182] ;

[0183] ;

[0184] Among them, h is a hidden layer variable, and h t is the input received from the previous node (i.e., the hidden layer variable at time t), and h t+1 is the output passed to the next node (i.e., the hidden layer variable at time t-1). Y is the output of the current state, and W h , W i , W o are different weight matrices, and the two sigmas are different linear connection layers. In the embodiments of the present application, h t can represent all historical information at time t and before time t, and h t-1 can represent all historical information at time t-1 and before time t-1.

[0185] Specifically, as shown in Figure 3 , for example, and input to the RNN recurrent neural network to obtain and , and input to the RNN recurrent neural network to obtain (the state information corresponding to the target vehicle at the current moment) and , and input to the RNN recurrent neural network to obtain and ;

[0186] Figure 3 In Figure 4 , D is the intention grid sampled from around the target vehicle. The sampling method is as shown in Figure 4 . The black dots in

[0187] represent the grid points on a dataset. Since the interval is too small, which is 1 foot, when sampling, 3×3 grid points are combined into a large grid, that is, the size of each grid is 4 feet × 4 feet. The coordinates of each grid are represented by the coordinates of the center point, and the intention label value is the average value of 9 points. Figure 4 As shown in

[0188] , a wider sampling is performed for the short distance in front of the vehicle, so that the model is more sensitive to the movement of the vehicle in the vertical lane direction, and a farther sampling is performed for the vehicle traveling direction, and the straight-ahead situation of the vehicle is more considered in this direction. Figure 3 In , φ is . Among them, and are two linear encoding modules, whose function is to encode the inputs and into the same dimension. D is a 9-row and 3-column matrix obtained by linking 9 intention grid points. The square brackets [] represent concat connection, and the two encoded matrices are connected as the input of the MLP. Among them, MLP is a multi-layer perceptron, is the intention matrix, representing the intention of the vehicle at this time. Among them, , .

[0189] Figure 3 The softmax module in , MLP is a multi-layer perceptron, is a 9-row and 3-column matrix. The 3 columns are the confidence levels of left turn, straight ahead, and right turn respectively. The softmax is used to obtain the scores of each possible case, specifically as Figure 5 shown, Through the MLP, is obtained, and then the softmax is used to obtain R t .

[0190] Specifically, through Z t and through the following formula, R t is obtained:

[0191] ;

[0192] Among them, is the result score of each intention, and the maximum value is taken as the intention result predicted by this grid, that is, .

[0193] Furthermore, in order to reduce the computational pressure during the intention prediction process and improve the accuracy of intention prediction, before predicting the state information corresponding to the target vehicle at the current moment by passing the position coordinates corresponding to each historical trajectory point through a recurrent neural network, it may further include: converting the position coordinates corresponding to each historical trajectory point from global coordinates to local coordinates; and converting the position coordinates corresponding to the target vehicle at the current moment from global coordinates to local coordinates. In the embodiments of the present application, if the position coordinates corresponding to the t-th moment (that is, the position coordinates corresponding to the current moment) are determined, the position coordinates at historical moments such as the t-1-th moment and the t-2-th moment... may have been converted from global coordinates to local coordinates, and h t-1, at this time, it is only necessary to convert the position coordinates corresponding to the t moment from the global coordinates to the local coordinates. It is also possible that the target vehicle does not perform vehicle intention prediction at historical moments. At this time, the position coordinates corresponding to the t moment (that is, the position coordinates corresponding to the current moment) and the position coordinates at historical moments can both be converted from the global coordinates to the local coordinates.

[0194] Specifically, taking the coordinates of the observation point with the earliest time among the historical trajectory observation points of the target vehicle as the origin, obs is the observation duration of the historical trajectory, the x-axis extends to the right, and the y-axis extends upward. Then, through the following formula, the position coordinates of the historical trajectory points are converted from the global coordinates to the local coordinates:

[0195] ;

[0196] where is the global coordinate, is the converted local coordinate, is the origin coordinate.

[0197] Furthermore, on the basis of converting the position coordinates corresponding to each historical trajectory point from the global coordinates to the local coordinates respectively, and converting the position coordinates corresponding to the target vehicle at the current moment to the local coordinates, the position coordinates corresponding to each historical trajectory point are predicted through a recurrent neural network to obtain the state information corresponding to the target vehicle at the current moment. Specifically, it can include: predicting the state information corresponding to the target vehicle at the current moment by passing the position coordinates corresponding to the target vehicle at the current moment and the local coordinates corresponding to each historical trajectory point through a recurrent neural network.

[0198] Furthermore, after predicting the state information corresponding to the target vehicle at each historical moment by passing the local coordinates corresponding to each historical trajectory point through a recurrent neural network, the position information, intention information, and the state information corresponding to the target vehicle at the current moment of each grid point in the grid where the target vehicle is currently located are used, and through a multi-layer perceptron, the vehicle intention at the current moment is predicted. Specifically, it can include: encoding the state information corresponding to the target vehicle at the current moment to obtain the encoded state information; encoding the position information and intention information of each grid point in the grid where the target vehicle is currently located to obtain the encoded grid information; connecting the encoded state information and the encoded grid information to obtain the connected encoded information; predicting the vehicle intention at the current moment based on the connected encoded information through a multi-layer perceptron.

[0199] Specifically, in the embodiments of the present application, the method of predicting the local coordinates corresponding to each historical trajectory point and the local coordinates corresponding to the current moment through a recurrent neural network can be found in the above embodiments and will not be elaborated here.

[0200] Further, before using the RNN and MLP used for intention prediction in the above embodiments to predict the vehicle intention at the current moment, the RNN and MLP are trained. Specifically, the cross-entropy loss function can be used as the loss function, as follows:

[0201] ;

[0202] Among them, 1 / 9 represents taking the average value of 9 grids in total. is the sign function (0 or 1), which is 1 if the sum of sample i is the same as the true intention, and 0 if it is different, that is, only the negative logarithm in the case of all correct predictions is calculated.

[0203] The true intention uses the intention label result of the vehicle in the dataset at this time as the true intention. By training, the loss function is made as small as possible so that the module can obtain a result closer to the true intention. The actual training module includes 3 weight matrices in the RNN encoding module, 2 linear encoding parts in φ, the weights of the MLP multi-layer perceptron, and the MLP module for dimensionality reduction before softmax. The final output result of this module is the intention matrix .

[0204] Further, through the above scheme, the vehicle intention of the target vehicle at the current moment can be obtained, that is, E t , of course, the vehicle intention of the target vehicle at the t+1 moment, the vehicle intention of the target vehicle at the t+2 moment, ……, the vehicle intention of the target vehicle at the t+N moment can also be obtained in the same way as the above scheme. The specific implementation method will not be elaborated in the embodiments of the present application. In the embodiments of the present application, the vehicle intention of the target vehicle at the t+1 moment, the vehicle intention of the target vehicle at the t+2 moment, ……, the vehicle intention of the target vehicle at the t+N moment, etc. are used for predicting the trajectory of the target vehicle.

[0205] Further, after obtaining the vehicle intention of the target vehicle through the above embodiments, that is, predicting the vehicle intention of the target vehicle based on the intention grid corresponding to the target vehicle, the following steps may also be included: obtaining the position coordinates of the trajectory points of the target vehicle at the current moment; obtaining the historical trajectory features of the target vehicle; obtaining the interaction features of the target vehicle at the current moment; predicting the vehicle trajectory information of the target vehicle based on the position coordinates of the trajectory points of the target vehicle at the current moment, the historical trajectory features of the target vehicle, and the interaction features at the current moment. Among them, the interaction features of the target vehicle at the current moment are used to characterize the interaction relationship between the target vehicle and other vehicles at the current moment; the historical trajectory features of the target vehicle are used to characterize the historical trajectory information of the target vehicle. In the embodiments of the present application, the historical trajectory features of the target vehicle may be obtained before obtaining the interaction features of the target vehicle, or the historical trajectory features of the target vehicle may be obtained after obtaining the interaction features of the target vehicle. Of course, the interaction features of the target vehicle may also be obtained while obtaining the historical trajectory features of the target vehicle.

[0206] Further, before obtaining the historical trajectory features of the target vehicle, the following steps may also be included: obtaining the historical trajectory information of the target vehicle; performing encoding processing on the position coordinates corresponding to each historical trajectory point to obtain the historical trajectory features of the target vehicle. In the embodiments of the present application, after obtaining the position information corresponding to each historical trajectory point of the target vehicle, encoding processing is performed through a recurrent neural network to obtain the historical trajectory features of the target vehicle.

[0207] Among them, the historical trajectory information is composed of multiple historical trajectory points.

[0208] Further, before obtaining the interaction features of the target vehicle at the current moment, the following steps may also be included: obtaining the position information corresponding to all vehicles at the current moment; selecting the position information of the preset vehicle from the position information corresponding to all vehicles at the current moment; determining the lane information where each selected vehicle is located at the current moment; generating the interaction features of the target vehicle at the current moment based on the position information of the preset vehicle and the lane information where each selected vehicle is located at the current moment. In the embodiments of the present application, the trajectory dataset in the above embodiments contains the trajectory information of all vehicles, and the historical trajectory information of all vehicles in the scene can be obtained from the above trajectory dataset. If the number of vehicles for which historical trajectory information is obtained is less than the preset number, a zero-padding operation is performed. If the number of vehicles for which historical trajectory information is obtained is greater than the preset number, vehicles closer to the target vehicle are preferentially selected as inputs, and vehicles farther away are not used as inputs, so as to obtain the historical trajectory information corresponding to the preset number of vehicles, and then generate the interaction features of the target vehicle based on the historical trajectory information corresponding to the obtained preset number of vehicles and their respective lane information. In the embodiments of the present application, the lane information is used to characterize the lane where the vehicle is currently located.

[0209] Specifically, generating the interaction features of the target vehicle at the current moment based on the preset vehicle position information and the lane information of each selected vehicle at the current moment includes: connecting the position information of each vehicle at the current moment with the lane information where each vehicle is located at the current moment; encoding the connected vehicle information to obtain the interaction features of the target vehicle at the current moment.

[0210] Specifically, as Figure 6 where XA represents the historical trajectories of all vehicles in the scenario, and , which is used to represent the historical trajectories corresponding to n vehicles at time t. Inputting the historical trajectories of the vehicles in the scenario, obs represents the length of the observed historical trajectories, where is the position of the i-th vehicle at time t, and the matrix consists of n rows and 2 columns, where n is the number of vehicles in this scenario;

[0211] represents the lane where the vehicle is located at this time. In the lane, the one closest to the yellow line is 1, and the one closest to the road edge is the maximum value. If there are 3 lanes on this road, the left lane is 1, the middle lane is 2, and the right lane is 3. Then, connect the lane where each vehicle is located at time t with its corresponding historical trajectory information, as shown in the formula where concat represents connecting and , represents the situation of the lane where the i-th vehicle is located in the scenario at time t. The connected matrix is n rows and 3 columns, as shown in the following formula:

[0212] ;

[0213] The connected result is sent to a multi-layer perceptron for encoding to increase the dimension of the data. The function of the MLP is the same at each time step here. Each time an RNN loop calculates the next time step, the same MLP is used, that is , the data is increased to k dimensions, and the data matrix is n rows and k columns at this time. Then, use a recurrent neural network (RNN) to encode the interaction data, as shown in the formula = , ).

[0214] Among them, the specific implementation method of the RNN is shown in the following formula:

[0215] = ∙ + ∙ ;

[0216] Among them, is the hidden layer variable, is the output passed to the previous node, is the input of the current state, , are different weight matrices, and σ is a linear connection layer.

[0217] Among them, the hidden layer variable represents the driving trajectories of all vehicles in the scenario. To consider the interaction between vehicles, the interaction between different trajectories will be considered in the following combination module. The final output result of this module is , indicating the interaction relationship between vehicles in the scenario, that is, the interaction feature of the target vehicle at time t.

[0218] Specifically, in the embodiment of the present application, the trajectory information of the target vehicle at the current moment and the historical trajectory features of the target vehicle are used to generate h t through RNN. After generating h t , the vehicle trajectory point q t of the target vehicle at time t + 1 can be predicted through h t+1 , and E t corresponding to the target vehicle at time t + 1 can be predicted through h t+1 , so as to further predict the vehicle trajectory point q t+2 of the target vehicle at time t + 2, and so on to obtain the vehicle trajectory points q t+N at time t + 3, time t + 4... time t + N.

[0219] Specifically, based on the trajectory point of the target vehicle at the current moment, the historical trajectory features of the target vehicle, and the interaction feature at the current moment, the vehicle trajectory information of the target vehicle is predicted. Specifically, it may include: step Sa (not shown in the figure) and step Sb (not shown in the figure), where

[0220] Step Sa: Perform trajectory prediction processing on the historical trajectory features of the target vehicle and the trajectory point of the target vehicle at the current moment to obtain the trajectory prediction result at the next moment.

[0221] For the embodiment of the present application, h t-1 of the target vehicle can be obtained based on the historical trajectory features of the target vehicle. Among them, the method for obtaining h t-1 of the target vehicle based on the historical trajectory features of the target vehicle can be seen in the above embodiment and will not be elaborated here.

[0222] Specifically, trajectory prediction processing is performed on the historical trajectory features of the target vehicle and the trajectory points of the target vehicle at the current moment to obtain the trajectory prediction result at the next moment, which may specifically include: encoding the historical trajectory features of the target vehicle and the trajectory points of the target vehicle at the current moment to obtain the hidden layer variable at the current moment; encoding based on the hidden layer variable at the current moment to obtain the trajectory prediction result at the next moment.

[0223] As Figure 7 shown, X t is the position information of the historical trajectory point of the target vehicle at time t, that is , where are the abscissa and ordinate of the vehicle at time t respectively. The trajectory points are input into the RNN recurrent neural network, and the RNN is used to encode the historical trajectory, that is

[0224] The specific implementation method of RNN is shown in the following formula;

[0225]

[0226] Among them, is the hidden layer variable, the input from the previous node received. X is the input of the current state, and W h , W i are different weight matrices. σ is the linear connection layer.

[0227] Furthermore, as Figure 7 shown, after obtaining h t-1 of the target vehicle, based on h t-1 and the trajectory point X t of the target vehicle at the current moment, and obtaining h t through the RNN, then based on h t and S t , and determining the trajectory point at time t + 1 (that is q t+1 ) through the RNN, and can also be obtained. Among them , S t+1 , S t+2 … are the results after encoding of the predicted points. When operating for the first time the initial value is set to a zero matrix.

[0228] Step Sb: For each target moment after the next moment in time, the following steps (Step 1 and Step 2) are sequentially executed in chronological order until the obtained trajectory prediction result meets the preset conditions.

[0229] Among them, the preset condition is to obtain the trajectory prediction result at a preset moment. For example, if the preset condition is to obtain the predicted trajectory at the moment of t+5, the loop ends after obtaining the moments of t+1, t+2, t+3, t+4, and t+5.

[0230] Step 1: Predict the vehicle intention at the previous moment of the target moment based on the trajectory prediction result at the previous moment of the target moment.

[0231] That is to say, the target moments are the moments of t+2, t+3, …, t+N. Among them, the previous moment of the moment of t+2 is the moment of t+1, the previous moment of the moment of t+3 is the moment of t+2, …, and the previous moment of the moment of t+N is the moment of t+(N-1).

[0232] Among them, for the first target moment, the trajectory prediction result at the previous moment of the target moment is the trajectory prediction result at the next moment. Further, for the first target moment (the moment of t+2), the trajectory prediction result at the previous moment of the target moment (the trajectory prediction result at the moment of t+1) is q t+1 。

[0233] Further, if the target moment is the moment of t+2, it is necessary to predict the vehicle intention at the moment of t+1 according to the trajectory prediction result q t+1 , that is, E t+1 ; if the target moment is the moment of t+3, it is necessary to predict the vehicle intention at the moment of t+2 according to the trajectory prediction result q t+2 , that is, E t+2 ; …, if the target moment is the moment of t+N, it is necessary to predict the vehicle intention at the moment of t+(N-1) according to the trajectory prediction result q t+（N-1） , that is, E t+（N-1） ; in the embodiments of the present application, the trajectory prediction result q t+1 at the moment of t+1, the trajectory prediction result q t+2 at the moment of t+2, …, the trajectory prediction result q t+（N-1） at the moment of t+(N-1) are used to predict E t+1 , E t+2 , …, E t+（N-1） respectively. The specific method is as described in the above vehicle intention prediction method and will not be elaborated here.

[0234] Step 2: Perform trajectory prediction processing based on the trajectory prediction result at the previous moment of the target moment, the interaction feature of the target vehicle at the current moment, and the vehicle intention at the previous moment of the target moment to obtain the trajectory prediction result at the target moment.

[0235] Specifically, in step 2, trajectory prediction processing is performed based on the trajectory prediction result at the previous moment of the target moment, the interaction features of the target vehicle at the current moment, and the vehicle intention at the previous moment of the target moment to obtain the trajectory prediction result at the target moment. Specifically, it may include: encoding through an attention mechanism based on the trajectory prediction result at the previous moment of the target moment, the interaction features of the target vehicle at the current moment, and the vehicle intention at the previous moment of the target moment to obtain the encoding result of the prediction point at the previous moment of the target moment; performing trajectory prediction processing based on the encoding result of the prediction point to obtain the trajectory prediction result at the target moment.

[0236] As Figure 7 shown, to obtain the trajectory prediction result q at time t+2 t+2 , the interaction features of the target vehicle at the current moment , the vehicle intention E of the target vehicle at time t+1 t+1 , are used to obtain S through the attention mechanism t+1 , and S t+1 and are used to obtain the trajectory prediction result q at time t+2 through RNN t+2 , and ; to obtain the trajectory prediction result q at time t+3 t+3 , the interaction features of the target vehicle at the current moment , the vehicle intention E of the target vehicle at time t+2 t+2 , are used to obtain S through the attention mechanism t+2 , and S t+2 and are used to obtain the trajectory prediction result q at time t+3 through RNN t+3 ; further, based on the same method as obtaining the trajectory prediction result q at time t+2 t+2 and the trajectory prediction result q at time t+3 t+3 , the trajectory prediction result q at time t+4 t+4 , ……, up to the trajectory prediction result q at time t+N t+N can be obtained.

[0237] Further, on the basis of obtaining the historical trajectory features of the target vehicle and the interaction features of the target vehicle through the above embodiments, and also obtaining the intention grid corresponding to the target vehicle currently, based on the position coordinates of the trajectory points of the target vehicle at the current moment, the historical trajectory features of the target vehicle, and the interaction features at the current moment, predict the vehicle trajectory information of the target vehicle. Specifically, it may include: performing trajectory prediction processing on the historical trajectory features of the target vehicle and the trajectory points of the target vehicle at the current moment to obtain the trajectory prediction result at the next moment; for each target moment after the next moment, sequentially perform the following steps (step 1 and step 2) in chronological order until the obtained trajectory prediction result meets the preset conditions, where,

[0238] Step 1, predict the vehicle intention at the previous moment of the target moment based on the trajectory prediction result at the previous moment of the target moment. For the first target moment, the trajectory prediction result at the previous moment of the target moment is the trajectory prediction result at the next moment;

[0239] Step 2, perform trajectory prediction processing based on the trajectory prediction result at the previous moment of the target moment, the interaction features of the target vehicle at the current moment, and the vehicle intention at the previous moment of the target moment to obtain the trajectory prediction result at the target moment.

[0240] Specifically, performing trajectory prediction processing on the historical trajectory features of the target vehicle and the trajectory points of the target vehicle at the current moment to obtain the trajectory prediction result at the next moment includes: performing encoding processing according to the historical trajectory features of the target vehicle and the trajectory points of the target vehicle at the current moment to obtain the hidden layer variable at the current moment; performing encoding processing based on the hidden layer variable at the current moment to obtain the trajectory prediction result at the next moment.

[0241] Specifically, performing trajectory prediction processing based on the trajectory prediction result at the previous moment of the target moment, the interaction features of the target vehicle at the current moment, and the vehicle intention at the previous moment of the target moment to obtain the trajectory prediction result at the target moment may specifically include: performing encoding through an attention mechanism based on the trajectory prediction result at the previous moment of the target moment, the interaction features of the target vehicle at the current moment, and the vehicle intention at the previous moment of the target moment to obtain the encoding result of the prediction point at the previous moment of the target moment; performing trajectory prediction processing based on the encoding result of the prediction point to obtain the trajectory prediction result at the target moment.

[0242] Specifically, based on the predicted vehicle intention, the historical trajectory features of the target vehicle, and the interaction features of the target vehicle, the vehicle trajectory information of the target vehicle is predicted, which may specifically include: performing trajectory prediction processing on the historical trajectory features of the target vehicle to obtain the trajectory prediction result at the next moment; determining the trajectory prediction result at the moment after the next moment based on the trajectory prediction result at the next moment, the interaction features of the target vehicle, and the vehicle intention predicted at the current moment; repeatedly determining the trajectory prediction result at the moment after the next moment based on the trajectory prediction result at the moment after the next moment, the interaction features of the target vehicle, and the vehicle intention predicted at the next moment until a preset condition is met. In the embodiments of the present application, the preset condition is to obtain the trajectory prediction result at the preset moment. For example, the preset condition is to obtain the predicted trajectory at the t+5 moment. Then, when the predictions at the t+1, t+2, t+3, t+4, and t+5 moments are obtained, the loop ends.

[0243] Specifically, determining the trajectory prediction result at the moment after the next moment based on the trajectory prediction result at the next moment, the interaction features of the target vehicle, and the vehicle intention predicted at the current moment may specifically include: encoding based on the trajectory prediction result at the next moment, the interaction features of the target vehicle, and the vehicle intention predicted at the current moment through an attention mechanism to obtain the encoding result of the prediction point at the next moment; performing trajectory prediction processing on the encoding result predicted at the next moment to obtain the trajectory prediction result at the moment after the next moment.

[0244] That is, as Figure 7 shown, after obtaining the historical trajectory encoding result , is passed to the RNN decoding module to obtain the trajectory prediction result at the t+1 moment, that is , where is the result after encoding the prediction point, and the initial value of is set to a zero matrix during the first operation.

[0245] The specific decoding process is as follows:

[0246] ;

[0247] The specific implementation manner of the RNN is as follows:

[0248] = ;

[0249] ;

[0250] The specific decoding process can also be as follows:

[0251] ;

[0252] The specific implementation manner of the RNN is as follows:

[0253] = ;

[0254] ;

[0255] where n ≥ 2, is a hidden layer variable, is the input received from the previous node. is the input of the current state, is the trajectory prediction result, W is different weight matrices, and two σ are different linear connection layers.

[0256] Furthermore, the interaction information and the intention information E and the historical trajectory information of the target vehicle, i.e., the hidden layer variable , and the output of the previous node are concatenated and then the interaction relationship between them is calculated using the Attention module.

[0257]

[0258]

[0259]

[0260] where is the weight matrix of the query vector, key vector, and value vector, is a hyperparameter, which is default set to 8, is the encoding result of the interaction information, is the vehicle intention information at time t + 1.

[0261] After that, the Attention calculation result will be used as the input for the next RNN decoder.

[0262] Through the RNN decoder, at time t, the predicted points from time t + 1 to t + N can be decoded based on the historical information to . At the next time step t + 1, N predicted points can be obtained again. When the RNN decoder runs for the first time the result is a zero matrix. When decoding with the RNN at the next time step, the obtained in the first step of the previous time step can be used as the input. The predicted points to obtained at each step will be used as the new input for intention prediction, i.e., the predicted value will be used as the input for intention prediction and the intention matrix will be updated to 。

[0263] For example, as Figure 7 shown, the historical trajectory features of the target vehicle at the current moment can be represented by and and are decoded by the RNN network to obtain and that is, the trajectory prediction result at time t + 1 and the hidden layer variable corresponding to time t + 1 are obtained. Then and are passed through the Attention and RNN networks to obtain and that is, the trajectory prediction result at time t + 2 and the hidden layer variable corresponding to time t + 2 are obtained. Then and are passed through the Attention and RNN networks to obtain and that is, the trajectory prediction result at time t + 1 and the hidden layer variable corresponding to time t + 3 are obtained. The above process is repeated until a preset condition is met. It should be noted that: S t can be a zero matrix.

[0264] Specifically, as Figure 7 shown, when and are passed through the Attention and RNN networks to obtain and , it specifically includes: , , and are used by Attention to obtain , and and are passed through the RNN network to obtain and ; and are passed through the Attention and RNN networks to obtain and , which specifically includes: , , and are used by Attention to obtain , and and are passed through the RNN network to obtain and 。

[0265] Further, through the above-mentioned Attention and RNN, the trajectory prediction results at each moment are obtained. The loss function for training Attention and RNN is as follows:

[0266]

[0267] The loss function uses the L2 distance difference between the predicted points and the ground truth trajectory, and calculates the average value of the distance differences of all predicted points as the loss function. In the embodiments of the present application, each time the model obtains the historical trajectory and the prediction result from the dataset, and then calculates the Loss with the corresponding predicted position. By changing the parameters in the model, the Loss is continuously reduced, that is, the distance difference between the model predicted points and the real points is continuously reduced, making the model prediction more and more accurate.

[0268] Further, in the above embodiments, after predicting the vehicle intention of the target vehicle, vehicle trajectory prediction is performed based on the vehicle intention of the target vehicle, thereby improving the accuracy of trajectory prediction. The trajectory prediction task is of great significance for improving the safety of driverless driving, which is still an important issue that needs to be solved in the driverless industry.

[0269] For some automated vehicles at levels L2 to L3, predicting the vehicle's intention is crucial for safe driving and the intelligent improvement and effective application of advanced driver assistance systems. The driver assistance system plays an important role in preventing traffic accidents and improving driving comfort, such as the blind spot monitoring system, lane departure warning system, and vehicle forward collision warning system.

[0270] The blind spot monitoring system, also known as the lane change assist system, mainly functions to eliminate the blind spot of the rearview mirror. It detects overtaking vehicles in the blind spots on both sides of the vehicle's rearview mirror through microwave radar and reminds the driver, thus avoiding accidents due to the blind spot of the rearview mirror during lane change. By predicting the vehicle intention, when the predicted vehicle intention is to change lanes, other vehicles can be detected, and reminders can be made only when needed, improving driving comfort.

[0271] The lane departure warning system is a system that assists the driver in reducing traffic accidents caused by the vehicle deviating from the lane by means of alarm. When it detects that the vehicle deviates from the lane, the sensor will collect vehicle data and the driver's operation status in a timely manner, and then the controller will issue an alarm signal. The whole process takes about 0.5 seconds to provide the driver with more reaction time. If the driver turns on the turn signal and changes lanes normally, then the lane departure warning system will not give any prompts. By predicting the vehicle intention, early warning can be achieved, thereby reducing the time required to issue the alarm signal, providing the driver with more reaction time, and thus reducing the risk of traffic accidents for the vehicle.

[0272] The forward collision warning system monitors the vehicle ahead at all times through a radar system, judges the distance, azimuth and relative speed between the vehicle itself and the vehicle ahead, and warns the driver when there is a potential collision risk. By predicting the intentions of surrounding vehicles, it can anticipate the lane-changing, turning and other behaviors of surrounding vehicles, and warn the driver when there is a potential collision risk, thereby reducing the risk of collision.

[0273] The above embodiments introduce a method for predicting vehicle intentions from the perspective of the method flow. The following embodiments introduce a device for predicting vehicle intentions from the perspective of virtual modules. For details, see the following embodiments.

[0274] An embodiment of the present application provides a device for predicting vehicle intentions, as Figure 8 shown. The device 80 for predicting vehicle intentions may specifically include: a first acquisition module 81, a vehicle intention prediction module 82, and a first generation module 83. Among them,

[0275] The first acquisition module 81 is configured to acquire an intention grid corresponding to a target vehicle;

[0276] The vehicle intention prediction module 82 is configured to predict the vehicle intention of the target vehicle based on the intention grid corresponding to the target vehicle;

[0277] Among them, when the first generation module 83 generates an intention grid, it is specifically configured to:

[0278] Acquire a trajectory data set and each scene map, and acquire the historical vehicle trajectories corresponding to each scene from the trajectory data set. The historical vehicle trajectories include the historical trajectories corresponding to multiple vehicles respectively, and the historical trajectory corresponding to any vehicle is composed of multiple historical trajectory points;

[0279] Screen the historical trajectory points within the preset area in each scene map from the historical vehicle trajectories corresponding to each scene respectively, and determine the intention information corresponding to each vehicle passing through each historical trajectory point;

[0280] Based on the intention information corresponding to each vehicle passing through each historical trajectory point, determine the intention information corresponding to each historical trajectory point;

[0281] Based on the intention information corresponding to each historical trajectory point and the position information corresponding to each historical trajectory point respectively, generate the intention grids corresponding to each scene respectively. The intention grid contains the position information corresponding to each historical trajectory point respectively and the intention information corresponding to each of them.

[0282] In a possible implementation manner of the embodiment of the present application, the device 80 further includes: a second acquisition module. Among them,

[0283] A second acquisition module, configured to acquire historical trajectory information of a target vehicle;

[0284] Wherein, when the vehicle intention prediction module 82 performs vehicle intention prediction on the target vehicle based on the intention grid corresponding to the target vehicle, it is specifically configured to:

[0285] Perform vehicle intention prediction on the target vehicle based on the historical trajectory information of the target vehicle and the intention grid corresponding to the target vehicle.

[0286] In another possible implementation manner of the embodiments of the present application, when the vehicle intention prediction module 82 performs vehicle intention prediction on the target vehicle based on the historical trajectory information of the target vehicle and the intention grid, it is specifically configured to:

[0287] Predict the vehicle intention at the current moment based on the historical trajectory information of the target vehicle, the intention grid, a recurrent neural network, and a multi-layer perceptron.

[0288] In another possible implementation manner of the embodiments of the present application, the historical trajectory information of the target vehicle is composed of multiple historical trajectory points, the intention grid includes multiple grids, and any grid includes multiple grid points;

[0289] When the vehicle intention prediction module 82 performs vehicle intention prediction at the current moment based on the historical trajectory information of the target vehicle, the intention grid, a recurrent neural network, and a multi-layer perceptron, it is specifically configured to:

[0290] Obtain the position coordinates corresponding to the target vehicle at the current moment;

[0291] Predict the state information corresponding to the target vehicle at the current moment by inputting the position coordinates corresponding to the target vehicle at the current moment and the position coordinates corresponding to each historical trajectory point into a recurrent neural network;

[0292] Based on the intention grid, determine the position information and intention information corresponding to each historical trajectory point in the grid where the target vehicle is currently located;

[0293] Predict the vehicle intention at the current moment by inputting the position information and intention information corresponding to each grid point, and the state information corresponding to the target vehicle at the current moment, into a multi-layer perceptron.

[0294] In another possible implementation manner of the embodiments of the present application, the apparatus 80 further includes: a conversion module, wherein,

[0295] The conversion module is configured to convert the position coordinates corresponding to each historical trajectory point from global coordinates to local coordinates; and,

[0296] Convert the position coordinates corresponding to the target vehicle at the current moment from global coordinates to local coordinates;

[0297] Among them, when the vehicle intention prediction module 82 predicts the position coordinates corresponding to the target vehicle at the current moment and the position coordinates corresponding to each historical trajectory point through a recurrent neural network to obtain the state information corresponding to the target vehicle at the current moment, it is specifically used for:

[0298] Predict the local coordinates corresponding to the target vehicle at the current moment and the local coordinates corresponding to each historical trajectory point through a recurrent neural network to obtain the state information corresponding to the target vehicle at the current moment.

[0299] In another possible implementation manner of the embodiment of the present application, when the vehicle intention prediction module 82 predicts the vehicle intention at the current moment by using the position information and intention information corresponding to each historical trajectory point in the grid where the target vehicle is currently located and the state information corresponding to the target vehicle at each historical moment, and through a multi-layer perceptron, it is specifically used for:

[0300] Encode the state information corresponding to the target vehicle at the current moment to obtain the encoded state information;

[0301] Encode the position information and intention information corresponding to each grid point in the grid where the target vehicle is currently located to obtain the encoded grid information;

[0302] Connect the encoded state information and the encoded grid information to obtain the connected encoded information;

[0303] Predict the vehicle intention at the current moment based on the connected encoded information and through a multi-layer perceptron.

[0304] In another possible implementation manner of the embodiment of the present application, the device 80 further includes: a third acquisition module, a fourth acquisition module, a fifth acquisition module, and a vehicle trajectory prediction module, where,

[0305] The third acquisition module is used to acquire the position coordinates of the trajectory points of the target vehicle at the current moment;

[0306] The fourth acquisition module is used to acquire the historical trajectory features of the target vehicle;

[0307] The fifth acquisition module is used to acquire the interaction feature of the target vehicle at the current moment, and the interaction feature of the target vehicle at the current moment is used to characterize the interaction relationship between the target vehicle and other vehicles at the current moment;

[0308] The vehicle trajectory prediction module is used to predict the vehicle trajectory information of the target vehicle based on the position coordinates of the trajectory points of the target vehicle at the current moment, the historical trajectory features of the target vehicle, and the interaction feature at the current moment.

[0309] Another possible implementation of the embodiment of the present application, the apparatus 80 further includes: a sixth acquisition module, a selection module, a determination module, and a generation module, where,

[0310] The sixth acquisition module is configured to acquire the position information corresponding to all vehicles at the current moment;

[0311] The selection module is configured to select the position information of a preset vehicle from the position information corresponding to all vehicles at the current moment;

[0312] The determination module is configured to determine the lane information in which each selected vehicle is located at the current moment;

[0313] The generation module is configured to generate the interaction feature of the target vehicle at the current moment based on the position information of the preset vehicle and the lane information in which each selected vehicle is located at the current moment.

[0314] Another possible implementation of the embodiment of the present application, when the generation module generates the interaction feature of the target vehicle at the current moment based on the position information of the preset vehicle and the lane information in which each selected vehicle is located at the current moment, specifically:

[0315] Connect the position information of each vehicle at the current moment with the lane information in which each vehicle is located at the current moment;

[0316] Encode the connected vehicle information to obtain the interaction feature of the target vehicle at the current moment.

[0317] Another possible implementation of the embodiment of the present application, the apparatus 80 further includes: a seventh acquisition module and an encoding processing module, where,

[0318] The seventh acquisition module is configured to acquire the historical trajectory information of the target vehicle, and the historical trajectory information is composed of multiple historical trajectory points;

[0319] The encoding processing module is configured to encode the position coordinates corresponding to each historical trajectory point to obtain the historical trajectory feature of the target vehicle.

[0320] Another possible implementation of the embodiment of the present application, when the vehicle trajectory prediction module predicts the vehicle trajectory information of the target vehicle based on the trajectory points of the target vehicle at the current moment, the historical trajectory feature of the target vehicle, and the interaction feature at the current moment, specifically:

[0321] Perform trajectory prediction processing on the historical trajectory feature of the target vehicle and the trajectory points of the target vehicle at the current moment to obtain the trajectory prediction result at the next moment;

[0322] For each target time after the next moment, the following steps are sequentially executed in chronological order until the obtained trajectory prediction result meets the preset conditions:

[0323] Predict the vehicle intention at the previous moment of the target time based on the trajectory prediction result at the previous moment of the target time. For the first target time, the trajectory prediction result at the previous moment of the target time is the trajectory prediction result at the next moment;

[0324] Perform trajectory prediction processing based on the trajectory prediction result at the previous moment of the target time, the interaction feature of the target vehicle at the current time, and the vehicle intention at the previous moment of the target time to obtain the trajectory prediction result at the target time.

[0325] In another possible implementation manner of the embodiment of the present application, when the vehicle trajectory prediction module performs trajectory prediction processing on the historical trajectory feature of the target vehicle and the trajectory point of the target vehicle at the current time to obtain the trajectory prediction result at the next moment, it specifically is used for:

[0326] Perform encoding processing according to the historical trajectory feature of the target vehicle and the trajectory point of the target vehicle at the current time to obtain the hidden layer variable at the current time;

[0327] Perform encoding processing based on the hidden layer variable at the current time to obtain the trajectory prediction result at the next moment.

[0328] In another possible implementation manner of the embodiment of the present application, when the vehicle trajectory prediction module performs trajectory prediction processing based on the trajectory prediction result at the previous moment of the target time, the interaction feature of the target vehicle at the current time, and the vehicle intention at the previous moment of the target time to obtain the trajectory prediction result at the target time, it specifically is used for:

[0329] Perform encoding through an attention mechanism based on the trajectory prediction result at the previous moment of the target time, the interaction feature of the target vehicle at the current time, and the vehicle intention at the previous moment of the target time to obtain the encoding result of the prediction point at the previous moment of the target time;

[0330] Perform trajectory prediction processing based on the encoding result of the prediction point to obtain the trajectory prediction result at the target time.

[0331] An embodiment of the present application provides a device for predicting vehicle intentions. Compared with the related art, in the embodiment of the present application, by obtaining the historical vehicle trajectories corresponding to each scenario map and screening out the historical trajectory points within the preset area in each scenario map, and then further determining the intention information corresponding to each vehicle passing through each trajectory point, the intention information corresponding to each trajectory point can be counted. Thus, according to the position information and intention information of each trajectory point, corresponding intention grids for each scenario are generated, so that when predicting the intention of a target vehicle, the intention grid corresponding to the target vehicle can be used to achieve vehicle intention prediction.

[0332] Those skilled in the art can clearly understand that for the convenience and brevity of description, the specific working processes of the above-described systems, devices, and units can refer to the corresponding processes in the foregoing method embodiments and will not be elaborated herein.

[0333] An embodiment of the present application provides an electronic device, such as Figure 9 shown. Figure 9 The electronic device 900 shown includes: a processor 901 and a memory 903. Among them, the processor 901 and the memory 903 are connected, such as through a bus 902. Optionally, the electronic device 900 may further include a transceiver 904. It should be noted that in actual applications, the transceiver 904 is not limited to one, and the structure of the electronic device 900 does not constitute a limitation to the embodiment of the present application.

[0334] The processor 901 may be a CPU (Central Processing Unit, central processor), a general-purpose processor, a DSP (Digital Signal Processor, data signal processor), an ASIC (Application Specific Integrated Circuit, application-specific integrated circuit), an FPGA (Field Programmable Gate Array, field programmable gate array), or other programmable logic devices, transistor logic devices, hardware components, or any combination thereof. It can implement or execute various exemplary logic blocks, modules, and circuits described in conjunction with the disclosure of the present application. The processor 901 may also be a combination for implementing computing functions, such as a combination including one or more microprocessors, a combination of a DSP and a microprocessor, etc.

[0335] The bus 902 may include a path for transmitting information between the above components. The bus 902 can be a PCI (Peripheral Component Interconnect) bus, an EISA (Extended Industry Standard Architecture) bus, or the like. The bus 902 can be divided into an address bus, a data bus, a control bus, etc. For ease of representation, Figure 9 only a thick line is used in Figure 9 , but it does not mean that there is only one bus or one type of bus.

[0336] The memory 903 can be a ROM (Read Only Memory) or other types of static storage devices that can store static information and instructions, a RAM (Random Access Memory) or other types of dynamic storage devices that can store information and instructions, or it can also be an EEPROM (Electrically Erasable Programmable Read Only Memory), a CD-ROM (Compact Disc Read Only Memory), or other optical disc storage, optical disc storage (including compact discs, laser discs, optical discs, digital versatile discs, Blu-ray discs, etc.), magnetic disk storage media, or other magnetic storage devices, or any other medium that can be used to carry or store the desired program code in the form of instructions or data structures and can be accessed by a computer, but is not limited thereto.

[0337] The memory 903 is used to store the application program code for implementing the solution of this application, and is controlled by the processor 901 to execute. The processor 901 is used to execute the application program code stored in the memory 903 to implement the content shown in the foregoing method embodiments.

[0338] Among them, the electronic device includes but is not limited to: mobile terminals such as mobile phones, laptop computers, digital broadcast receivers, PDAs (Personal Digital Assistants), PADs (Tablet Computers), PMPs (Portable Multimedia Players), in-vehicle terminals (such as in-vehicle navigation terminals), etc., and fixed terminals such as digital TVs, desktop computers, etc. It can also be a server, etc. Figure 9 The electronic device shown is only an example and should not impose any limitations on the functions and usage scope of the embodiments of this application.

[0339] An embodiment of the present application provides a computer-readable storage medium. A computer program is stored on the computer-readable storage medium. When it runs on a computer, it enables the computer to execute the corresponding content in the foregoing method embodiment. Compared with the related art, in the embodiment of the present application, by obtaining the historical vehicle trajectories corresponding to each scene map respectively, and screening out the historical trajectory points belonging to the preset area in each scene map, and then further determining the intention information corresponding to each vehicle passing through each trajectory point, the intention information corresponding to each trajectory point can be counted. Thus, according to the position information and intention information of each trajectory point, corresponding intention grids corresponding to each scene are generated, so that when predicting the intention of a target vehicle, the intention grid corresponding to the target vehicle can be based on to achieve vehicle intention prediction.

[0340] Those skilled in the art can clearly understand that for the convenience and simplicity of description, only the above division of each functional module is used as an example. In actual applications, the above functions can be allocated to different functional modules according to needs, that is, the internal structure of the device is divided into different functional modules to complete all or part of the functions described above. The specific working processes of the above-described system, device, and unit can refer to the corresponding processes in the foregoing method embodiment and will not be repeated here.

[0341] In the embodiments provided in the present application, it should be understood that the disclosed system, device, and method can be implemented in other ways. For example, the device embodiments described above are merely illustrative. For example, the division of the module or unit is only a logical function division. In actual implementation, there may be other division methods. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the displayed or discussed mutual coupling or direct coupling or communication connection can be through some interfaces, and the indirect coupling or communication connection of the device or unit can be in an electrical, mechanical or other form.

[0342] The unit described as a separate component may or may not be physically separated, and the component displayed as a unit may or may not be a physical unit, that is, it may be located in one place, or may be distributed to multiple network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0343] In addition, each functional unit in each embodiment of the present application can be integrated in a processing unit, or each unit exists physically alone, or two or more units are integrated in one unit. The above integrated unit can be implemented in the form of hardware or in the form of a software functional unit.

[0344] When the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or part of this technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) or a processor to execute all or part of the steps of the methods described in various embodiments of this application. The aforementioned storage medium includes: various media that can store program codes, such as USB flash drives, mobile hard disks, read-only memories, random access memories, magnetic disks, or optical discs.

[0345] As described above, the above embodiments are only used to introduce the technical solution of this application in detail. However, the description of the above embodiments is only for helping to understand the method and its core idea of this application, and should not be construed as a limitation of this application. Any changes or substitutions that can be easily thought of by those skilled in the art within the technical scope disclosed in this application should be covered within the protection scope of this application.

Claims

1. A method for predicting vehicle intention, characterized in that, Including: Obtain the intention grid corresponding to the target vehicle; Based on the intention grid corresponding to the target vehicle, perform vehicle intention prediction on the target vehicle; Among them, the generation method of the intention grid includes: Obtain the trajectory dataset and each scenario map, and obtain the historical vehicle trajectories corresponding to each scenario from the trajectory dataset. The historical vehicle trajectories include the historical trajectories corresponding to multiple vehicles respectively, and the historical trajectory corresponding to any vehicle is composed of multiple historical trajectory points; Screen the historical trajectory points within the preset area in each scenario map from the historical vehicle trajectories corresponding to each scenario respectively, and determine the intention information corresponding to each vehicle passing through each historical trajectory point; Based on the intention information corresponding to each vehicle passing through each historical trajectory point, determine the intention information corresponding to each historical trajectory point; Based on the intention information corresponding to each historical trajectory point and the position information corresponding to each historical trajectory point respectively, generate the intention grid corresponding to each scenario. The intention grid contains the position information corresponding to each historical trajectory point and the intention information corresponding to each of them; Before performing vehicle intention prediction on the target vehicle based on the intention grid corresponding to the target vehicle, it also includes: Obtain the historical trajectory information of the target vehicle; Among them, performing vehicle intention prediction on the target vehicle based on the intention grid corresponding to the target vehicle includes: Based on the historical trajectory information of the target vehicle and the intention grid corresponding to the target vehicle, perform vehicle intention prediction on the target vehicle; Performing vehicle intention prediction on the target vehicle based on the historical trajectory information of the target vehicle and the intention grid corresponding to the target vehicle includes: Based on the historical trajectory information of the target vehicle and the intention grid, and through a recurrent neural network and a multi-layer perceptron, predict the vehicle intention at the current moment; the historical trajectory information of the target vehicle is composed of multiple historical trajectory points, and the intention grid contains multiple grids, and any grid contains multiple grid points; Based on the historical trajectory information of the target vehicle and the intention grid, and through a recurrent neural network and a multi-layer perceptron, predicting the vehicle intention at the current moment includes: Obtain the position coordinates of the target vehicle at the current moment; Predict the position coordinates corresponding to the target vehicle at the current moment and the position coordinates corresponding to each historical trajectory point respectively through a recurrent neural network to obtain the state information of the target vehicle at the current moment; Based on the intention grid, determine the position information and intention information corresponding to each grid point in the grid where the target vehicle is currently located; Predict the vehicle intention at the current moment through a multi-layer perceptron with the position information and intention information corresponding to each grid point respectively, and the state information of the target vehicle at the current moment.

2. The method according to claim 1, characterized in that Before predicting the state information of the target vehicle at the current moment by predicting the position coordinates corresponding to the target vehicle at the current moment and the position coordinates corresponding to each historical trajectory point respectively through a recurrent neural network, it also includes: Convert the position coordinates corresponding to each historical trajectory point from global coordinates to local coordinates; and, Convert the position coordinates corresponding to the target vehicle at the current moment from global coordinates to local coordinates; Among them, the position coordinates corresponding to the target vehicle at the current moment and the position coordinates corresponding to each historical trajectory point are predicted through a recurrent neural network to obtain the state information corresponding to the target vehicle at the current moment, including: Predict the position coordinates corresponding to the target vehicle at the current moment and the position coordinates corresponding to each historical trajectory point through a recurrent neural network to obtain the state information corresponding to the target vehicle at the current moment.

3. The method according to claim 1, characterized in that, Predict the vehicle intention at the current moment by using the position information and intention information corresponding to each historical trajectory point in the grid where the target vehicle is currently located and the state information corresponding to the target vehicle at the current moment, and through a multi-layer perceptron, including: Perform encoding processing on the state information corresponding to the target vehicle at the current moment to obtain the encoded state information; Perform encoding processing on the position information and intention information corresponding to each grid point in the grid where the target vehicle is currently located to obtain the encoded grid information; Connect the encoded state information and the encoded grid information to obtain the connected encoded information; Predict the vehicle intention at the current moment based on the connected encoded information and through the multi-layer perceptron.

4. The method according to any one of claims 1 to 3, characterized in that, The method further includes: Obtain the position coordinates of the trajectory points of the target vehicle at the current moment; Obtain the historical trajectory features of the target vehicle; Obtain the interaction features of the target vehicle at the current moment, where the interaction features of the target vehicle at the current moment are used to characterize the interaction relationship between the target vehicle and other vehicles at the current moment; Predict the vehicle trajectory information of the target vehicle based on the position coordinates of the trajectory points of the target vehicle at the current moment, the historical trajectory features of the target vehicle, and the interaction features at the current moment.

5. The method according to claim 4, characterized in that Before obtaining the interaction features of the target vehicle at the current moment, it further includes: Obtain the position information corresponding to each vehicle at the current moment; Select the position information of the preset vehicle from the position information corresponding to each vehicle at the current moment; Determine the lane information where each selected vehicle is located at the current moment; Generate the interaction features of the target vehicle at the current moment based on the position information of the preset vehicle and the lane information where each selected vehicle is located at the current moment.

6. The method according to claim 5, characterized in that The generating the interaction features of the target vehicle at the current moment based on the position information of the preset vehicle and the lane information where each selected vehicle is located at the current moment includes: Connect the position information of each vehicle at the current moment with the lane information where each vehicle is located at the current moment; Perform encoding processing on the connected vehicle information to obtain the interaction features of the target vehicle at the current moment.

7. The method according to any one of claims 5-6, characterized in that, Before obtaining the historical trajectory features of the target vehicle, it further includes: Obtain the historical trajectory information of the target vehicle, where the historical trajectory information is composed of multiple historical trajectory points; Encode the position coordinates corresponding to each historical trajectory point to obtain the historical trajectory feature of the target vehicle.

8. The method according to claim 4, characterized in that Predicting the vehicle trajectory information of the target vehicle based on the trajectory point of the target vehicle at the current moment, the historical trajectory feature of the target vehicle, and the interaction feature at the current moment includes: Performing trajectory prediction processing on the historical trajectory feature of the target vehicle and the trajectory point of the target vehicle at the current moment to obtain a trajectory prediction result at the next moment; For each target moment after the next moment, perform the following steps in sequence according to the time sequence until the obtained trajectory prediction result meets the preset conditions: Predict the vehicle intention at the previous moment of the target moment based on the trajectory prediction result at the previous moment of the target moment. For the first target moment, the trajectory prediction result at the previous moment of the target moment is the trajectory prediction result at the next moment; Perform trajectory prediction processing based on the trajectory prediction result at the previous moment of the target moment, the interaction feature of the target vehicle at the current moment, and the vehicle intention at the previous moment of the target moment to obtain the trajectory prediction result at the target moment.

9. The method according to claim 8, characterized in that The performing trajectory prediction processing on the historical trajectory feature of the target vehicle and the trajectory point of the target vehicle at the current moment to obtain a trajectory prediction result at the next moment includes: Performing encoding processing according to the historical trajectory feature of the target vehicle and the trajectory point of the target vehicle at the current moment to obtain a hidden layer variable at the current moment; Performing encoding processing based on the hidden layer variable at the current moment to obtain the trajectory prediction result at the next moment.

10. The method according to claim 8 or 9, characterized in that The performing trajectory prediction processing based on the trajectory prediction result at the previous moment of the target moment, the interaction feature of the target vehicle at the current moment, and the vehicle intention at the previous moment of the target moment to obtain the trajectory prediction result at the target moment includes: Encoding through an attention mechanism based on the trajectory prediction result at the previous moment of the target moment, the interaction feature of the target vehicle at the current moment, and the vehicle intention at the previous moment of the target moment to obtain an encoding result of the prediction point at the previous moment of the target moment; Performing trajectory prediction processing based on the encoding result of the prediction point to obtain the trajectory prediction result at the target moment.

11. An apparatus for predicting vehicle intention, characterized in that, Including: A first acquisition module for acquiring an intention grid corresponding to the target vehicle; A vehicle intention prediction module for predicting the vehicle intention of the target vehicle based on the intention grid corresponding to the target vehicle; Wherein, when the first generation module generates the intention grid, it is specifically used for: Acquiring a trajectory data set and each scenario map, and acquiring the historical vehicle trajectories corresponding to each scenario from the trajectory data set. The historical vehicle trajectories include the historical trajectories corresponding to multiple vehicles respectively, and the historical trajectory corresponding to any vehicle is composed of multiple historical trajectory points; Screening the historical trajectory points in the preset area of each scenario map from the historical vehicle trajectories corresponding to each scenario respectively, and determining the intention information corresponding to each vehicle passing through each historical trajectory point; Determine the intention information corresponding to each historical trajectory point based on the intention information corresponding to each vehicle passing through each historical trajectory point; Generate an intention grid corresponding to each scenario based on the intention information corresponding to each historical trajectory point and the position information corresponding to each historical trajectory point, where the intention grid includes the position information corresponding to each historical trajectory point and its corresponding intention information; Before predicting the vehicle intention of the target vehicle based on the intention grid corresponding to the target vehicle, further include: Obtain the historical trajectory information of the target vehicle; Among them, predicting the vehicle intention of the target vehicle based on the intention grid corresponding to the target vehicle includes: Predict the vehicle intention of the target vehicle based on the historical trajectory information of the target vehicle and the intention grid corresponding to the target vehicle; Predicting the vehicle intention of the target vehicle based on the historical trajectory information of the target vehicle and the intention grid corresponding to the target vehicle includes: Predict the vehicle intention at the current moment based on the historical trajectory information of the target vehicle and the intention grid and through a recurrent neural network and a multi-layer perceptron; The historical trajectory information of the target vehicle is composed of multiple historical trajectory points, the intention grid includes multiple grids, and any grid includes multiple grid points; Predicting the vehicle intention at the current moment based on the historical trajectory information of the target vehicle and the intention grid and through a recurrent neural network and a multi-layer perceptron includes: Obtain the position coordinates of the target vehicle at the current moment; Predict the state information of the target vehicle at the current moment by passing the position coordinates of the target vehicle at the current moment and the position coordinates of each historical trajectory point through a recurrent neural network; Based on the intention grid, determine the position information and intention information corresponding to each grid point in the grid where the target vehicle is currently located; Predict the vehicle intention at the current moment by passing the position information and intention information corresponding to each grid point, and the state information of the target vehicle at the current moment, through a multi-layer perceptron.

12. An electronic device, characterized in that, Include: One or more processors; A memory; One or more applications, where the one or more applications are stored in the memory and configured to be executed by the one or more processors, and the one or more applications are configured to: execute a method for predicting vehicle intention according to any one of claims 1 to 10.

13. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by a processor, it implements a method for predicting vehicle intention according to any one of claims 1 to 10.

Images