Vehicle trajectory reconstruction method and device

Abstract

The embodiment of the invention provides a vehicle track reconstruction method and device, and relates to the technical field of intelligent traffic. Comprising the steps of obtaining an initial vehicle track of a target vehicle, road network data of an area where the target vehicle is located and signal lamp state data; inputting the initial vehicle track, the road network data and the signal lamp state data into a target data driving reconstruction model to obtain a plurality of target track points corresponding to the initial vehicle track; determining a reconstructed vehicle track corresponding to the initial vehicle track based on the initial vehicle track and the plurality of target track points; the target data driving reconstruction model obtains a plurality of target trajectory points through the following steps: extracting vehicle motion features in an initial vehicle trajectory; carrying out data fusion on the signal lamp state data and the vehicle motion characteristics, and predicting a plurality of track points of the target vehicle based on the fused data; and constraining the plurality of track point coordinates based on the road network data to obtain a plurality of target track points conforming to the road network data.

Description

Technical Field

[0001] The present application relates to the field of intelligent transportation technology, and in particular to a vehicle trajectory reconstruction method and device. Background Art

[0002] Vehicle trajectories contain a wealth of traffic information. They not only reflect the vehicle's driving path, but also contain a wealth of vehicle spatiotemporal information, and are the most intuitive reflection of vehicle behavior. However, detection equipment will be disturbed by noise during data collection, processing, and transmission. Trajectory data will have shortcomings such as uncertainty, sparsity, and skewed distribution. It is impossible to accurately project vehicle trajectories on the real road environment, which limits its application in the field of transportation. Reconstructing vehicle trajectories is a prerequisite for data detection and further application.

[0003] There are two main types of trajectory reconstruction methods: mechanism-driven and data-driven. The mechanism model is constrained by the function form and is difficult to characterize the vehicle motion state in complex scenarios. The data-driven model only relies on a large amount of complete trajectory data for training, lacks consideration of real application scenarios, reduces the model's applicability, and makes the accuracy of the reconstructed vehicle trajectory predicted by the model poor. Summary of the invention

[0004] The present application provides a vehicle trajectory reconstruction method and device for improving the accuracy of the predicted reconstructed vehicle trajectory.

[0005] A first aspect of an embodiment of the present application provides a vehicle trajectory reconstruction method, including: obtaining an initial vehicle trajectory of a target vehicle, road network data and signal light state data of an area where the target vehicle is located; inputting the initial vehicle trajectory, road network data and signal light state data into a target data-driven reconstruction model, and obtaining a plurality of target trajectory points corresponding to the initial vehicle trajectory through the target data-driven reconstruction model; determining a reconstructed vehicle trajectory corresponding to the initial vehicle trajectory based on the initial vehicle trajectory and the plurality of target trajectory points; the target data-driven reconstruction model obtains the plurality of target trajectory points through the following steps: extracting vehicle motion features from the initial vehicle trajectory; fusing the signal light state data with the vehicle motion features, and predicting a plurality of trajectory points of the target vehicle based on the fused data; constraining the coordinates of the plurality of trajectory points based on the road network data, and obtaining a plurality of target trajectory points that conform to the road network data.

[0006] The vehicle trajectory reconstruction method provided in the embodiment of the present application predicts the reconstructed vehicle trajectory of the target vehicle by inputting the initial vehicle trajectory of the target vehicle, the road network data of the area where the target vehicle is located, and the signal light state data into the target data-driven reconstruction model. It is understandable that the signal light state data provides the real-time status of the traffic lights (such as red light, green light, yellow light) and its timing information. Combined with these data, the behavior of the vehicle at the intersection, such as parking, accelerating, decelerating or passing, can be predicted more accurately. This helps to consider the influence of the signal light on the vehicle's speed, direction and time when reconstructing the trajectory. The road network data can simulate the driving conditions of the vehicle under different road conditions, further improve the accuracy of trajectory prediction, and at the same time, the road network data can also help identify possible driving paths, thereby providing more reasonable choices when reconstructing the trajectory.

[0007] In other words, the road network data can constrain the position of the trajectory points of the vehicle trajectory, and the traffic light state data is related to the time corresponding to the trajectory points of the vehicle trajectory. Then, based on the road network data and traffic light state data of the area where the target vehicle is located, the initial vehicle trajectory of the target vehicle is reconstructed. Taking into account the impact of real application scenarios on the vehicle trajectory, the predicted reconstructed vehicle trajectory is more accurate.

[0008] In combination with a first implementation method of the first aspect, obtaining an initial vehicle trajectory of a target vehicle includes: obtaining an original vehicle trajectory of the target vehicle; identifying, among multiple trajectory points of the original vehicle trajectory, abnormal trajectory points whose trajectory points do not conform to a road range to which the trajectory points belong; and deleting the abnormal trajectory points in the original vehicle trajectory to obtain an initial vehicle trajectory.

[0009] In combination with the second implementation method of the first aspect, before the initial vehicle trajectory, road network data and signal light state data are input into the target data-driven reconstruction model, the method also includes: dividing the initial vehicle trajectory into multiple trajectory segments; identifying whether there are abnormal trajectory segments in the multiple trajectory segments; the abnormal trajectory segments have one or more of the following characteristics: the existence of corresponding position points with a residence time higher than a preset threshold, multiple trajectory points with the same position, trajectory jitter or trajectory offset; wherein the trajectory jitter is used to indicate that the running direction angle of the target vehicle changes multiple times within a preset time period, and the trajectory offset is used to indicate that the distance between multiple trajectory points in the abnormal trajectory segment and the center line of the lane to which the trajectory points belong exceeds a preset distance; deleting the abnormal trajectory segments in the initial vehicle trajectory.

[0010] In combination with the third implementation method of the first aspect, the target data-driven reconstruction model includes an encoder and a decoder, the encoder is used to extract vehicle motion features in the initial vehicle trajectory; the decoder includes a trajectory point prediction layer and a predicted position stack, the trajectory point prediction layer is used to fuse signal light state data with vehicle motion features, and predict multiple trajectory points of the target vehicle based on the fused data; the predicted position stack is used to constrain the coordinates of multiple trajectory points based on road network data to obtain multiple target trajectory points that conform to the road network data.

[0011] In combination with the fourth implementation method of the first aspect, the target data-driven reconstruction model is trained in the following manner: obtaining multiple training samples, each training sample including: a sample vehicle trajectory, road network data and signal light state data corresponding to the sample vehicle trajectory; removing some trajectory points in the sample vehicle trajectory; inputting the sample vehicle trajectory after removing the trajectory points, the road network data and signal light state data corresponding to the sample vehicle trajectory into the initial data-driven reconstruction model to obtain multiple target trajectory points corresponding to the sample vehicle trajectory; calculating the first loss corresponding to the sample vehicle trajectory based on the partial trajectory points removed from the sample vehicle trajectory and the multiple target trajectory points corresponding to the sample vehicle trajectory; updating the initial data-driven reconstruction model by the first losses corresponding to the multiple sample vehicle trajectories until the iteration stop condition is met to obtain the target data-driven reconstruction model.

[0012] In combination with a fifth implementation of the first aspect, an initial data-driven reconstruction model is updated by first losses corresponding to multiple sample vehicle trajectories until an iteration stop condition is met to obtain a target data-driven reconstruction model, including: determining a position, speed, and preceding vehicle information of a sample vehicle corresponding to a first trajectory point in the sample vehicle trajectory; the first trajectory point is any trajectory point except the last trajectory point in the sample vehicle trajectory; the preceding vehicle information includes a position and speed of a preceding vehicle of the sample vehicle; the preceding vehicle of the sample vehicle is a vehicle that is in the same lane as the sample vehicle and is in front of the sample vehicle; the position, speed, and preceding vehicle information of the sample vehicle corresponding to the first trajectory point are input into an initial vehicle following model, and a predicted moving distance of the sample vehicle within a preset period after the time corresponding to the first trajectory point is predicted by the initial vehicle following model; an actual moving distance of the sample vehicle within a preset time after the time corresponding to the first trajectory point is determined based on the sample vehicle trajectory; a second loss corresponding to the sample vehicle trajectory is determined based on the predicted moving distances and the actual moving distances corresponding to the multiple first trajectory points; an initial fusion model is updated based on the first losses and the second losses corresponding to the multiple sample vehicle trajectories until the iteration stop condition is met to obtain a target data-driven reconstruction model, and the initial fusion model includes an initial data-driven reconstruction model and an initial vehicle following model.

[0013] In combination with the sixth implementation of the first aspect, the initial vehicle following model predicts the predicted moving distance of the sample vehicle in the following manner: determining the expected vehicle distance between the sample vehicle and the preceding vehicle based on the speed of the sample vehicle and the speed of the preceding vehicle; determining the distance between the sample vehicle and the preceding vehicle based on the position of the sample vehicle and the position of the preceding vehicle, and determining the vehicle acceleration of the sample vehicle at the first trajectory point based on the expected vehicle distance and the distance between the sample vehicle and the preceding vehicle; determining the predicted moving distance of the sample vehicle based on the vehicle acceleration of the sample vehicle at the first trajectory point, the speed of the sample vehicle, and the length of a preset time period; the expected vehicle distance is calculated by the following formula:

[0014]

[0015] Among them, S * is the expected vehicle distance; s0 is the minimum safe distance; T is the safe following distance; a is the maximum acceleration of the vehicle; b is the maximum deceleration of the vehicle; is the speed of the sample vehicle at the first trajectory point; is the speed of the preceding vehicle.

[0016] In combination with the seventh implementation method of the first aspect, determining the preceding vehicle information corresponding to the first trajectory point in the trajectory of the sample vehicle includes: determining multiple neighboring vehicles whose distance from the first trajectory point of the sample vehicle is within a preset distance threshold; determining the running angle of each neighboring vehicle among the multiple neighboring vehicles, and the running angle of the sample vehicle at the first trajectory point; among the multiple neighboring vehicles, deleting the neighboring vehicles whose running angle difference with the sample vehicle is greater than the preset angle difference, to obtain multiple target neighboring vehicles; among the multiple target neighboring vehicles, taking the target neighboring vehicle that is in the same lane as the sample vehicle and is located in front of the sample vehicle as the preceding vehicle of the sample vehicle; obtaining the position and speed of the preceding vehicle of the sample vehicle, to obtain the preceding vehicle information corresponding to the first trajectory point.

[0017] In combination with the eighth implementation method of the first aspect, the initial vehicle trajectory includes multiple trajectory points, each trajectory point corresponds to a trajectory point data, and the trajectory point data includes: collection time, longitude and latitude position, vehicle speed, running angle and lane of the vehicle; road network data includes intersections, sections, lanes and topological relationships on the road; signal light state data includes green light duration, red light duration and green light utilization rate corresponding to the lane where the target vehicle is located.

[0018] A second aspect of an embodiment of the present application provides a vehicle trajectory reconstruction device, including: an acquisition module, used to acquire an initial vehicle trajectory of a target vehicle, road network data and signal light state data of an area where the target vehicle is located; an input module, used to input the initial vehicle trajectory, road network data and signal light state data into a target data-driven reconstruction model, and obtain multiple target trajectory points corresponding to the initial vehicle trajectory through the target data-driven reconstruction model; a determination module, used to determine a reconstructed vehicle trajectory corresponding to the initial vehicle trajectory based on the initial vehicle trajectory and the multiple target trajectory points; the target data-driven reconstruction model obtains the multiple target trajectory points through the following steps: extracting vehicle motion features in the initial vehicle trajectory; fusing the signal light state data with the vehicle motion features, and predicting multiple trajectory points of the target vehicle based on the fused data; constraining the coordinates of the multiple trajectory points based on the road network data, and obtaining multiple target trajectory points that conform to the road network data.

[0019] A third aspect of an embodiment of the present application provides an electronic device, comprising: one or more processors; one or more memories; wherein the one or more memories are used to store computer program codes, the computer program codes include computer instructions, and when the one or more processors execute the computer instructions, the electronic device executes the vehicle trajectory reconstruction method provided by the first aspect and its possible implementation methods.

[0020] A fourth aspect of an embodiment of the present application provides a computer-readable storage medium, which stores computer execution instructions. When the computer execution instructions are executed on a computer, the computer executes the vehicle trajectory reconstruction method provided by the first aspect and its possible implementation methods.

[0021] A fifth aspect of an embodiment of the present application provides a computer program product, including a computer program / instruction, which, when executed by a processor, implements the steps of the vehicle trajectory reconstruction method provided in the first aspect and its possible implementation methods.

[0022] Among them, the beneficial effects described in the second to fifth aspects can refer to the beneficial effect analysis of the first aspect, and will not be repeated here. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] The accompanying drawings are used to provide a further understanding of the technical solution of the present invention and constitute a part of the specification. Together with the embodiments of the present application, they are used to explain the technical solution of the present invention and do not constitute a limitation on the technical solution of the present invention.

[0024] Figure 1 A method flow of a vehicle trajectory reconstruction method provided in an embodiment of the present application Figure 1 ;

[0025] Figure 2A schematic diagram of the structure of an encoder provided in an embodiment of the present application;

[0026] Figure 3 A schematic diagram of the structure of a decoder provided in an embodiment of the present application;

[0027] Figure 4 A method flow of a vehicle trajectory reconstruction method provided in an embodiment of the present application Figure 2 ;

[0028] Figure 5 A method flow chart of a target data driven reconstruction model training method provided in an embodiment of the present application;

[0029] Figure 6 A method flow of a vehicle trajectory reconstruction method provided in an embodiment of the present application Figure 3 ;

[0030] Figure 7 A schematic diagram of a grid-based process for identifying a preceding vehicle provided in an embodiment of the present application;

[0031] Figure 8 A method flow of a vehicle trajectory reconstruction method provided in an embodiment of the present application Figure 4 ;

[0032] Fig. 9 A schematic diagram of the structure of a vehicle trajectory reconstruction device provided in an embodiment of the present application;

[0033] Fig.10 A schematic diagram of the structure of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0034] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0035] The terms "first" and "second" are used for descriptive purposes only and should not be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of the features. In the description of this application, unless otherwise specified, "plurality" means two or more.

[0036] In the description of this application, it should be noted that, unless otherwise clearly specified and limited, the terms "connected" and "connection" should be understood in a broad sense, for example, it can be a fixed connection, a detachable connection, or an integral connection. For ordinary technicians in this field, the specific meanings of the above terms in this application can be understood according to specific circumstances. In addition, when describing a pipeline, the "connected" and "connection" used in this application have the meaning of conduction. The specific meaning needs to be understood in conjunction with the context.

[0037] In the embodiments of the present application, words such as "exemplary" or "for example" are used to indicate examples, illustrations or descriptions. Any embodiment or design described as "exemplary" or "for example" in the embodiments of the present application should not be interpreted as being more preferred or more advantageous than other embodiments or designs. Specifically, the use of words such as "exemplary" or "for example" is intended to present related concepts in a specific way.

[0038] Vehicle trajectories can reflect the temporal and spatial variation characteristics of the vehicle's instantaneous speed and are the most comprehensive and complete form of expression of the traffic flow operation status. Through vehicle trajectories, it is possible to identify the key points at which vehicles join and leave queues, queue assembly waves and dissipation waves, intersection arrival flow, etc., thereby providing data support for signal control optimization, etc. In an intelligent network environment, vehicle trajectory data may have missing segments due to data transmission, vehicle occlusion, sensor detection range, etc. Through trajectory reconstruction, these missing segments can be estimated and filled, improving the accuracy and completeness of the data, thereby comprehensively reflecting the traffic status and facilitating subsequent applications and analysis.

[0039] There are two main types of trajectory reconstruction methods: mechanism-driven and data-driven. The mechanism-driven method mainly reconstructs vehicle trajectories based on physical principles and traffic flow theory. This type of method usually uses the kinematic characteristics of the vehicle and the dynamic changes of traffic flow to simulate and predict the vehicle's driving trajectory through mathematical models and algorithms. The advantage of the mechanism-driven method is that it has a solid theoretical foundation and can reflect the essential characteristics of vehicle movement. However, due to the complexity and uncertainty of the actual traffic environment, the mechanism-driven method often requires simplified assumptions and conditions, which may lead to a certain deviation between the reconstruction results and the actual situation.

[0040] The data-driven model relies on a large amount of full-sequence trajectory data for training, and uses statistical analysis, machine learning, deep learning and other technical means to extract the vehicle's motion characteristics and laws in the trajectory data, and reconstruct the vehicle's driving trajectory accordingly. However, it lacks consideration of real application scenarios, which reduces the model's applicability and makes the accuracy of the reconstructed vehicle trajectory predicted by the model poor.

[0041] Based on this, the embodiment of the present application provides a vehicle trajectory reconstruction method, by inputting the initial vehicle trajectory of the target vehicle, the road network data of the area where the target vehicle is located, and the signal light state data, into the target data driven reconstruction model to predict multiple target trajectory points of the target vehicle, and then obtain the reconstructed vehicle trajectory based on the initial vehicle trajectory and multiple target trajectory points corresponding to the target vehicle. It can be understood that the road network data mainly includes information such as the basic attributes, geometric shapes, connection relationships, and traffic facilities of the road, and the signal light state data refers to the state information of the traffic lights at different time points. When the vehicle moves, it needs to conform to the basic shape of the road in the road network data, etc., and it is necessary to determine whether to stop or move forward based on the state information of the traffic lights. That is, the road network data can constrain the position of the trajectory points of the vehicle trajectory, and the signal light state data is related to the time corresponding to the trajectory points of the vehicle trajectory, and then the initial vehicle trajectory of the target vehicle is reconstructed based on the road network data and the signal light state data of the area where the target vehicle is located, taking into account the impact of the real application scenario on the vehicle trajectory, and the predicted reconstructed vehicle trajectory is more accurate.

[0042] It should be understood that the vehicle trajectory reconstruction method provided in the embodiment of the present application is applicable to a trajectory reconstruction device deployed with a target data-driven reconstruction model. Among them, the trajectory reconstruction device can be a terminal device, such as a personal computer (PC), a laptop computer, a mobile device, a tablet computer, a laptop computer, etc. The embodiment of the present application does not limit the specific form of the electronic device. Alternatively, the trajectory reconstruction device can also be a separate server, or a server cluster composed of multiple servers. In some implementations, the server cluster can be a distributed cluster server. The embodiment of the present application does not impose any restrictions on the specific form of the model training device.

[0043] See also Figure 1 , the vehicle trajectory reconstruction method provided in the embodiment of the present application includes S101-S103:

[0044] S101, obtaining an initial vehicle trajectory of a target vehicle, road network data of an area where the target vehicle is located, and traffic light state data.

[0045] It should be understood that the initial vehicle trajectory is a vehicle trajectory formed by multiple actual position points (trajectory points) of the target vehicle. Due to data transmission, vehicle occlusion, sensor detection range, etc., there may be missing trajectory points in the initial vehicle trajectory. Trajectory reconstruction is required to estimate and fill these missing trajectory points to obtain a complete reconstructed vehicle trajectory.

[0046] Exemplarily, as a feasible implementation method, the initial vehicle trajectory includes multiple trajectory points, each trajectory point corresponds to a trajectory point data, and the trajectory point data includes: collection time, longitude and latitude position, vehicle speed, running angle and lane of the vehicle.

[0047] Road network data mainly includes information such as the basic attributes, geometric shapes, connection relationships, and transportation facilities of roads. Road network data can be obtained through a variety of channels. For example, road network data can be obtained through map data providers, which provide detailed road network data, including basic information such as road grade, geometric shapes, slopes, curvatures, and directions; or road network data can be obtained through traffic management departments, which also have a large amount of road network information data, including information such as road construction, traffic control, and traffic accidents.

[0048] Exemplarily, as a feasible implementation method, the road network data includes intersections, road sections, lanes and topological relationships on the road.

[0049] Traffic light status data refers to the status information of traffic lights at different time points, which can include the duration of red, green, and yellow lights and the transition intervals between them. There are mainly the following ways to obtain traffic light status data: Traffic management center, which collects traffic light status data in real time through sensors and cameras installed on the road; Intelligent transportation system, some advanced intelligent transportation systems can obtain traffic light status data in real time and automatically adjust the timing of traffic lights according to traffic flow and vehicle driving conditions. These systems usually integrate advanced communication technology, sensor technology, and artificial intelligence algorithms to achieve intelligent control of traffic signals; On-board equipment, some on-board equipment (such as navigation devices, driving recorders, etc.) also have the function of obtaining traffic light status data. These devices can obtain real-time information on the status of the traffic lights in front of the vehicle through built-in positioning modules and image recognition technology.

[0050] Exemplarily, as a feasible implementation method, the signal light state data includes the green light duration, red light duration and green light utilization rate corresponding to the lane where the target vehicle is located.

[0051] S102, inputting the initial vehicle trajectory, the road network data and the signal light state data into the target data driven reconstruction model to obtain a plurality of target trajectory points corresponding to the initial vehicle trajectory.

[0052] Specifically, the target data driven reconstruction model provided in the embodiment of the present application obtains multiple target trajectory points corresponding to the initial vehicle trajectory by executing the following steps S201-S203:

[0053] S201: extracting vehicle motion features from an initial vehicle trajectory.

[0054] In some embodiments, the target data driven reconstruction model may include an encoder and a decoder, wherein the encoder is used to extract vehicle motion features in an initial vehicle trajectory, and the decoder is used to predict a reconstructed vehicle trajectory based on the vehicle motion features as well as road network data and signal light state data.

[0055] As a feasible implementation method, Figure 2 As shown, the encoder may include an embedding layer and a long short-term memory network (LSTM) layer. The embedding layer will include a linear layer and a LeakyReLU activation function, which is a commonly used combination for learning nonlinear representations of data.

[0056] Among them, the main function of the embedding layer is to convert high-dimensional sparse input data into low-dimensional dense vector representation. Specifically, the linear layer is used to linearly transform the input data into a new space. In this new space, the data may be easier to understand and process by the model. The linear layer performs weighted summation on the input through the weight matrix and bias term.

[0057] After the linear transformation, the activation function introduces nonlinearity, which enables the model to learn complex patterns. LeakyReLU is a variant of ReLU (Rectified Linear Unit) that has a small slope in the negative region, thus avoiding the "dead neuron" problem that ReLU may cause. This helps the model better capture the nonlinear characteristics of the data.

[0058] The LSTM layer is a variant of a recurrent neural network that is specifically designed to handle long-term dependencies in sequence data. In vehicle trajectory encoding, the LSTM layer is able to capture the dependencies between different time points in the trajectory, thereby learning the dynamic characteristics of the trajectory and obtaining the vehicle motion characteristics in the initial vehicle trajectory. This is very useful for predicting future trajectory points or identifying patterns in the trajectory.

[0059] The initial vehicle trajectory is processed through the above-mentioned encoder network structure including an embedding layer (composed of a linear layer and a LeakyReLU activation function) and an LSTM layer, and a series of features related to vehicle motion, namely, vehicle motion features, can be obtained.

[0060] As a feasible implementation method, the vehicle motion feature can include the summary information of all time steps in the trajectory. In the initial trajectory data of the vehicle, each trajectory point can be regarded as the observation result of a time step. That is, the trajectory point is the position or state of the object at a specific time step.

[0061] S202: Fusing the signal light state data with the vehicle motion characteristics, and predicting multiple trajectory points of the target vehicle based on the fused data.

[0062] The signal light state data is integrated with the vehicle motion features extracted in S201, which can be achieved by simple splicing or more complex feature extraction and encoding. It should be understood that since the signal light state data can affect the behavior of the vehicle, such as determining whether the vehicle needs to stop at the traffic light and how long it needs to stop and wait, the integrated feature set contains important factors that affect the vehicle behavior and is the basis for subsequent model prediction.

[0063] In some embodiments, the decoder in the reconstruction model can be driven by the target data to predict the coordinates of multiple trajectory points of the target vehicle based on the fused data. Figure 3 As shown, the decoder may include a track point prediction layer and a prediction position stack.

[0064] The trajectory point prediction layer is used to fuse the signal light state data with the vehicle motion characteristics, and predict multiple trajectory points of the target vehicle based on the fused data. The predicted position stack is used to constrain the coordinates of multiple trajectory points based on the road network data to obtain multiple target trajectory points that conform to the road network data.

[0065] Optional, such as Figure 3 As shown, the trajectory point prediction layer may include a multi-layer perceptron (MLP), an LSTM, and a linear layer.

[0066] As a feasible implementation method, the signal light state data can be fused with the vehicle motion features and then input into the MLP. The MLP can learn the complex interaction of features and output a low-dimensional feature representation, namely the historical position features of the vehicle. It should be understood that the purpose of dimensionality reduction is to reduce the complexity of the model, improve computational efficiency, and possibly remove redundant information.

[0067] The historical location features of the vehicle are then used as the input of LSTM. LSTM is a neural network that is particularly suitable for processing sequence data and can capture temporal dependencies. The LSTM layer will process the input historical location features in sequence and generate the location embedding features of the next moment. These location embedding features capture the movement trend and possible paths of the vehicle.

[0068] The processing process of the LSTM layer is repeatedly executed, each time predicting the position embedding features of the next time step based on the output of the previous time step, and iteratively predicting multiple time steps within the time period corresponding to the initial vehicle trajectory to obtain all predicted state features.

[0069] The predicted state features output by LSTM are input into the linear layer, which is responsible for mapping the high-dimensional predicted state features to two-dimensional spatial coordinates, that is, obtaining multiple trajectory points of the predicted target vehicle.

[0070] It should be noted that Figure 3 LSTM(128) means that the number of features output by the LSTM layer is 128, Linear(128) means that the number of features output by a linear layer is 128, and Linear(2) means that the number of features output by a linear layer is 2.

[0071] S203, constraining multiple trajectory points based on the road network data to obtain coordinates of multiple target trajectory points that conform to the road network data.

[0072] The predicted trajectory points are constrained and adjusted according to the road network data. This can ensure the feasibility and safety of the predicted trajectory. The constraint process may include removing illegal trajectory points (such as predicting areas outside the road), adjusting the trajectory to comply with traffic regulations, etc.

[0073] In some embodiments, Figure 3 As shown, the decoder also includes a predicted position stack, which is used to constrain the coordinates of multiple trajectory points of multiple target vehicles according to the road network data to obtain multiple target trajectory points. Among them, the predicted position stack can be a data structure for storing and managing the predicted positions (i.e., trajectory point coordinates) of multiple target vehicles at different time points in the future. This data structure may exist in the form of an array, a list, a matrix, or a more complex form, and is intended to uniformly manage and constrain multiple trajectories for subsequent analysis, optimization, and decision-making.

[0074] As a feasible implementation method, the predicted location stack can compare and constrain multiple trajectory points predicted by the trajectory point prediction layer with the road network data to ensure the rationality and feasibility of the predicted location. Specifically, this includes the following aspects:

[0075] Road connectivity: Trajectory points must be located on actual roads, and the roads must be connected. This requires the prediction algorithm to recognize the structure of the road network and ensure that the paths between the predicted trajectory points are reachable.

[0076] Traffic rules: Trajectory points need to comply with traffic rules, such as traffic lights, one-way streets, no turns, etc. These rules are usually encoded in the road network data, and the prediction algorithm needs to consider the restrictions of these rules during the prediction process.

[0077] Road geometry: The geometry of the road (such as curvature, width, slope, etc.) affects the trajectory of the moving object. The predicted position stack needs to use the road geometry information in the road network data to fine-tune and optimize the trajectory points.

[0078] S103: Determine a reconstructed vehicle trajectory corresponding to the initial vehicle trajectory based on the initial vehicle trajectory and multiple target trajectory points.

[0079] After obtaining the initial vehicle trajectory and multiple target trajectory points, the multiple target trajectory points can be inserted into the initial vehicle trajectory according to their corresponding time points to obtain a reconstructed vehicle trajectory corresponding to the initial vehicle trajectory.

[0080] It can be seen that the vehicle trajectory reconstruction method provided by the embodiment of the present application predicts multiple target trajectory points corresponding to the initial vehicle trajectory by inputting the initial vehicle trajectory of the target vehicle, the road network data of the area where the target vehicle is located, and the signal light state data into the target data-driven reconstruction model; and then based on the initial vehicle trajectory and multiple target trajectory points, the reconstructed vehicle trajectory of the target vehicle is obtained. It is understandable that the signal light state data provides the real-time status of traffic lights (such as red lights, green lights, yellow lights) and their timing information. Combined with these data, the behavior of vehicles at intersections, such as parking, accelerating, decelerating or passing, can be more accurately predicted. This helps to consider the impact of signal lights on vehicle speed, direction and time when reconstructing the trajectory. Road network data can simulate the driving conditions of vehicles under different road conditions, further improve the accuracy of trajectory prediction, and at the same time, road network data can also help identify possible driving paths, thereby providing more reasonable choices when reconstructing the trajectory.

[0081] In other words, the road network data can constrain the position of the trajectory points of the vehicle trajectory, and the traffic light state data is related to the time corresponding to the trajectory points of the vehicle trajectory. Then, based on the road network data and traffic light state data of the area where the target vehicle is located, the initial vehicle trajectory of the target vehicle is reconstructed. Taking into account the impact of real application scenarios on the vehicle trajectory, the predicted reconstructed vehicle trajectory is more accurate.

[0082] In some embodiments, since various errors often occur during the collection of vehicle trajectory data, noise data is often hidden in a large amount of trajectory data and is difficult to be found. Therefore, it is usually necessary to preprocess the original collected vehicle trajectory data before it can be used for subsequent predictions.

[0083] As a feasible implementation method, obtaining the initial vehicle trajectory of the target vehicle in S201 may include: obtaining the original vehicle trajectory of the target vehicle; identifying abnormal trajectory points among multiple trajectory points of the original vehicle trajectory whose trajectory points do not conform to the road range to which the trajectory points belong; and deleting the abnormal trajectory points in the original vehicle trajectory to obtain the initial vehicle trajectory.

[0084] It is understandable that if a track point is not within the polygonal range of the road to which it belongs, it indicates that there may be an error (positioning error or signal interference) when recording the track point, resulting in a deviation between the recorded track point position and the actual road position. This deviation may cause the track point to fall outside the road polygon range, that is, the position of the track point is inaccurate. Therefore, it is necessary to identify whether the track point conforms to the road range to which the track point belongs.

[0085] For each road, its geometry can be regarded as a polygon (for complex roads, it may need to be decomposed into multiple polygons), and then multiple trajectory points of the original vehicle trajectory are traversed. For each trajectory point, check whether it is within the polygon range of the road to which it belongs. If not, it can be marked as an abnormal trajectory point and removed from the original vehicle trajectory to filter out trajectory points that are beyond the road network range or do not match the road.

[0086] As a feasible implementation method, after deleting the abnormal trajectory points in the original vehicle trajectory, the filtering algorithm can be used to process the original vehicle trajectory, and the filtering and smoothing time window can be defined according to the trajectory reconstruction duration of the trajectory reconstruction task to eliminate the noise problem in the trajectory data. Finally, the vehicle trajectory after filtering is used as the initial vehicle trajectory of the target vehicle.

[0087] Filtering can effectively reduce the noise in trajectory data, which may come from sensor errors, environmental interference and other factors. Through filtering, the trajectory data can be closer to the actual motion state of the vehicle. The trajectory curve after filtering is smoother, avoiding the mutation points or outliers that may appear in the original data, thereby improving the continuity and consistency of the trajectory data. The trajectory data after filtering can more accurately reflect the movement trend and law of the vehicle, thus providing a more reliable basis for subsequent trajectory prediction.

[0088] It can be seen that when obtaining the initial vehicle trajectory of the vehicle, the original vehicle trajectory of the vehicle can be obtained first, and then the abnormal trajectory points in the original vehicle trajectory can be identified and deleted to filter out the trajectory points that exceed the road network range or do not match the road, which can ensure the accuracy of the initial vehicle trajectory and facilitate subsequent processing.

[0089] In some embodiments, the vehicle trajectory may include some abnormal trajectory segments, such as long parking time or trajectory circling, which may be caused by some unexpected situations of the driver and do not reflect the actual movement state of the vehicle. In the reconstruction process, if these outliers are not removed, they may be mistakenly taken into account, causing the reconstructed trajectory to deviate from the actual path. Therefore, before using the target data to drive the reconstruction model to predict the reconstructed vehicle trajectory of the target vehicle, it is necessary to delete the abnormal trajectory segments in the initial vehicle trajectory.

[0090] As a feasible implementation method, before inputting the initial vehicle trajectory, road network data and signal light state data into the target data-driven reconstruction model, refer to Figure 4 The vehicle trajectory reconstruction method provided in the embodiment of the present application further includes the following steps:

[0091] S401: Divide an initial vehicle trajectory into multiple trajectory segments.

[0092] In practical applications, vehicle trajectories may be divided in a variety of ways, which are not limited in the embodiments of the present application.

[0093] Exemplarily, as a feasible implementation, the initial vehicle trajectory may be divided into a plurality of trajectory segments based on a preset time interval, that is, a time threshold is set, such as dividing the initial vehicle trajectory into a plurality of trajectory segments according to the time threshold.

[0094] As another feasible implementation method, the trajectory is divided by analyzing the direction, angle change or distance of the trajectory points. If the direction angle change of the trajectory point exceeds a certain threshold, or the distance between adjacent trajectory points is too large, segmentation is performed; or key points in the trajectory are selected as segmentation points to reduce the redundant information of the trajectory while maintaining the main features of the trajectory; or the initial vehicle trajectory is divided into multiple trajectory segments according to a preset distance threshold.

[0095] S402: Identify whether there is an abnormal trajectory segment among the multiple trajectory segments.

[0096] The abnormal trajectory segment has one or more of the following characteristics: the existence of a corresponding position point whose residence time is higher than a preset threshold, multiple trajectory points with the same position, trajectory jitter or trajectory deviation.

[0097] If a trajectory segment has a corresponding location point with a dwell time higher than a preset threshold, it indicates that the target vehicle has stopped for a long time at that location point. Trajectory segments with long parking times may contain a large amount of stationary or low-speed moving data, which are noise data for analyzing the vehicle's mobility characteristics and trajectory patterns. Deleting these trajectory segments can reduce the noise in the data and improve the accuracy and reliability of subsequent analysis.

[0098] If a trajectory segment has multiple trajectory points with the same position, it indicates that the target vehicle has reversed or circled. The vehicle appears at the same position after multiple time intervals. The reason may be that the target vehicle has traveled to the wrong position and needs to adjust its position. Therefore, the trajectory segment is regarded as an abnormal trajectory segment.

[0099] Track jitter is used to indicate that the target vehicle's running direction angle changes multiple times within a preset time period, that is, the vehicle's running angle changes unnecessarily and unnaturally in a short period of time. This change may be caused by measurement errors, signal interference, equipment failure, or data recording errors. It should be understood that jittery track segments will distort the vehicle's actual driving path, making the track data inaccurate. Deleting these segments can restore the vehicle's actual driving trajectory and improve the accuracy of the data.

[0100] Track deviation is used to indicate that the distance between multiple track points in the abnormal track segment and the center line of the lane to which the track points belong exceeds the preset distance. It is understandable that if the target vehicle is operating normally, the distance between it and the center line of the lane to which it belongs is within the moving range. If track deviation occurs, it may be caused by factors such as unstable GPS signal, equipment error, and environmental interference. Deleting these segments can significantly reduce the noise level in the data, making the remaining data purer and more reliable.

[0101] S403: Delete the abnormal trajectory segment in the initial vehicle trajectory.

[0102] After the abnormal trajectory segments in the initial vehicle trajectory are identified, in order to improve the data quality of the initial vehicle trajectory, these abnormal trajectory segments are deleted to reduce the noise in the data and improve the purity and reliability of the remaining data.

[0103] The above embodiment describes the method of reconstructing the vehicle trajectory from the perspective of model application. In order to facilitate the explanation of the effect of the target data driven reconstruction model provided in the embodiment of the present application, the training process of the target data driven reconstruction model is described below.

[0104] It should be understood that the training process of the target data-driven reconstruction model can be applied to the trajectory reconstruction device in the above embodiment, and can also be applied to the model training device. Among them, the model training device can be a terminal device, such as a personal computer (PC), a laptop computer, a mobile device, a tablet computer, a laptop computer, etc., and the embodiment of the present application does not limit the specific form of the electronic device. Alternatively, the model training device can also be a separate server, or it can be a server cluster composed of multiple servers. In some implementations, the server cluster can be a distributed cluster server. The embodiment of the present application does not impose any restrictions on the specific form of the model training device.

[0105] See also Figure 5 The target data driven reconstruction model provided in the embodiment of the present application is obtained by training in the following manner:

[0106] S501: Obtain multiple training samples.

[0107] Each training sample includes: a sample vehicle trajectory, road network data corresponding to the sample vehicle trajectory, and signal light state data.

[0108] It should be understood that before model training, it is necessary to install positioning equipment (such as GPS receiver) and other sensors (such as speed sensor, acceleration sensor, etc.) on the vehicle to record the vehicle's driving trajectory and status information in real time. Due to equipment failure, signal interference, etc., there may be noise, outliers or missing values ​​in the original data. In the data processing stage, these data need to be cleaned to remove noise and outliers, such as pre-processing operations such as identifying abnormal trajectory points, filtering or deleting abnormal trajectory segments in the above embodiments, and then filter out qualified trajectory data from the processed data as sample vehicle trajectories. The screening conditions may include the integrity, length, time range, etc. of the trajectory.

[0109] S502: Remove some track points in the sample vehicle track.

[0110] Since the data-driven reconstruction model is needed to reconstruct the sample vehicle trajectory, but the relatively complete sample vehicle trajectory does not need to be reconstructed, and the data-driven reconstruction model cannot reconstruct it. Therefore, it is necessary to remove some trajectory points in the sample vehicle trajectory so that the data-driven reconstruction model can reconstruct the removed trajectory points.

[0111] S503, inputting the sample vehicle trajectory after removing the trajectory points, the road network data corresponding to the sample vehicle trajectory, and the signal light state data into the initial data-driven reconstruction model to obtain a plurality of target trajectory points corresponding to the sample vehicle trajectory.

[0112] The sample vehicle trajectory after the trajectory points are removed is input into the initial data-driven reconstruction model, and the removed trajectory points are reconstructed through the initial data-driven reconstruction model to obtain the reconstructed vehicle trajectory corresponding to the sample vehicle trajectory.

[0113] S504: Calculate a first loss corresponding to the sample vehicle trajectory based on a portion of trajectory points removed from the sample vehicle trajectory and a plurality of target trajectory points corresponding to the sample vehicle trajectory.

[0114] During the model training phase, the multiple target trajectory points corresponding to the sample vehicle trajectory predicted by the initial data-driven reconstruction model are somewhat different from the actual trajectory points of the sample vehicle. Therefore, it is necessary to calculate the first loss based on the removed trajectory points and the predicted multiple target trajectory points.

[0115] It should be understood that the embodiment of the present application does not limit the method of calculating the first loss. In practical applications, the mean squared error (MSE), mean absolute error (MAE) or dynamic time warping (DTW) can be used to calculate the first loss.

[0116] Among them, MSE is one of the most commonly used regression loss functions, which calculates the average of the square of the difference between the predicted value and the true value. For trajectory data, we can use the position of each time point in the trajectory data (usually longitude and latitude or two-dimensional coordinates) as the predicted value and the true value to calculate MSE.

[0117] MAE is another commonly used regression loss function that calculates the average of the absolute values ​​of the differences between the predicted values ​​and the true values. Compared with MSE, MAE is less sensitive to outliers.

[0118] If the reconstructed trajectory and the true trajectory are not completely aligned on the time axis, you can consider using DTW. DTW is a method to measure the similarity between two time series, which allows local stretching and bending on the time axis.

[0119] Exemplarily, as a feasible implementation method, the first loss can be calculated using the following formula (1):

[0120]

[0121] Where LAE represents the first loss, N is the number of trajectory points, and are the positions of the i-th points in the removed trajectory points, and are the positions of the i-th point in the reconstructed multiple target trajectory points.

[0122] S505 , updating the initial data-driven reconstruction model using the first losses corresponding to the plurality of sample vehicle trajectories until an iteration stop condition is met, thereby obtaining a target data-driven reconstruction model.

[0123] After obtaining the first loss corresponding to the sample vehicle trajectory, the gradient of each parameter in the model can be calculated through the back propagation algorithm based on the gradient information of the first loss. Then, the optimizer is used to update the parameters of the model according to the gradient of the parameters in the hope of reducing the loss. The steps of forward propagation, loss calculation, back propagation and parameter update are repeated through the first losses corresponding to multiple sample vehicle trajectories until the iteration stop condition is met.

[0124] It should be understood that the iteration stop condition can be set based on a variety of factors, including but not limited to the following conditions:

[0125] Reach maximum number of iterations: Set a fixed number of iterations as the upper limit of training.

[0126] Loss convergence: If the change in loss over multiple consecutive iterations is very small (below a certain threshold), the model is considered to have converged and training can be stopped.

[0127] Validation set performance: Use the validation set to evaluate the performance of the model. If the performance of the model on the validation set begins to deteriorate (i.e., overfitting occurs), stop training.

[0128] It can be seen that the target data-driven reconstruction model training method provided in this embodiment obtains multiple sample vehicle trajectories and corresponding road network data and traffic light state data for model training, so that the model can more accurately capture the characteristics and rules in the data, thereby improving the precision and accuracy of reconstruction, and has good generalization ability, and can maintain good performance on unseen data.

[0129] In some embodiments, the reconstructed trajectory of the vehicle not only needs to meet the constraints of the road network information and the signal light state data, but also needs to conform to certain vehicle kinematics knowledge. Based on this, the vehicle trajectory reconstruction method provided in the embodiment of the present application can combine vehicle kinematics knowledge with the data-driven method when reconstructing the vehicle trajectory, that is, the vehicle following model and the data-driven reconstruction model are integrated, so that when reconstructing the trajectory, not only can the vehicle motion characteristics be mined from the historical trajectory, but also the prediction results can be guaranteed to meet the vehicle motion constraints, thereby ensuring the generalization and accuracy of the model.

[0130] For details, please refer to Figure 6 , S505 can be specifically implemented as the following steps:

[0131] S601: Determine the position, speed, and preceding vehicle information of a sample vehicle corresponding to a first trajectory point in a sample vehicle trajectory.

[0132] The first trajectory point is any trajectory point in the sample vehicle trajectory except the last trajectory point; the preceding vehicle information includes the position and speed of the vehicle that is in the same lane as the sample vehicle and is in front of the sample vehicle.

[0133] It should be understood that the vehicle following model is a model that describes the corresponding reaction of the following vehicle (FV) caused by the change of the leading vehicle (LV) in the driving convoy on a single lane with restricted overtaking. It uses the dynamic method to study the corresponding behavior of the following vehicle caused by the change of the leading vehicle's motion state, and then predicts the moving distance of the current vehicle within a period of time.

[0134] Therefore, for any trajectory point in the sample vehicle trajectory, it is necessary to determine the preceding vehicle position information corresponding to the sample vehicle at the trajectory point, and then train the vehicle following model through the position, speed and preceding vehicle position information of the sample vehicle corresponding to the first trajectory point in the sample vehicle trajectory.

[0135] As a feasible implementation method, Figure 7 As shown in the figure, a plane rectangular coordinate system can be established with the sample vehicle as the origin, the vehicle driving direction as the longitudinal direction (Y), and the vertical driving direction as the transverse direction (X). Considering the influence of surrounding vehicles on the predicted trajectory of the sample vehicle, the road network is divided into unit grids of a specified size (such as 3*7) with the sample vehicle as the center, which is used to describe the positions of all vehicles within the research range at the same frame time. In the process of trajectory feature extraction, a feature dimension of the corresponding size (such as 3*7=21) is added.

[0136] For example, Figure 7 As shown, the first 7 columns represent the vehicle information of the left lane, the middle 7 columns represent the vehicle information of the current lane of the sample vehicle, and the last 7 columns represent the vehicle information of the right lane. Each cell is stored using the vehicle number, and the trajectory information of the neighboring vehicle can be queried later by using the number.

[0137] When the position information of the preceding vehicle needs to be obtained, the preceding vehicle of the sample vehicle can be obtained by querying the vehicle number, and then the position information of the preceding vehicle can be obtained.

[0138] S602: input the position, speed and preceding vehicle information of the sample vehicle corresponding to the first trajectory point into an initial vehicle following model, and predict the predicted moving distance of the sample vehicle within a preset period of time after the time corresponding to the first trajectory point through the initial vehicle following model.

[0139] The vehicle following model uses the dynamic method to study the vehicle's moving distance. Therefore, the vehicle following model needs to be affected by a series of dynamic parameters when applied, such as acceleration, acceleration change rate, free flow speed, acceleration control coefficient, etc., so that the vehicle following model can predict the vehicle's moving distance based on the dynamic method.

[0140] It should be understood that the preset period is a pre-set time interval, which can be set according to needs in the actual application process, and the embodiment of the present application does not limit this. Exemplarily, as a feasible implementation method, the preset period can be 5 minutes.

[0141] As a feasible implementation method, the vehicle following model predicts the predicted moving distance of the sample vehicle within a preset period after the time corresponding to the first trajectory point in the following way:

[0142] S11. Determine an expected vehicle distance between the sample vehicle and the preceding vehicle according to the speed of the sample vehicle and the speed of the preceding vehicle.

[0143] Specifically, as a feasible implementation method, the expected distance between the sample vehicle and the preceding vehicle can be calculated by the following formula (2):

[0144]

[0145] Among them, S * represents the expected vehicle distance; s0 is the minimum safe distance; T is the safe following distance; a is the maximum acceleration of the vehicle; b is the maximum deceleration of the vehicle; is the speed of the sample vehicle at the first trajectory point; is the speed of the preceding vehicle of the sample vehicle at the time corresponding to the first trajectory point.

[0146] It is understandable that the minimum safety distance is used to ensure that there is enough space between the two vehicles even when they are stationary. It includes the length of the vehicle itself and the safety buffer distance reserved for unexpected situations. This distance is fixed and does not change with the speed or speed difference, but it provides a basic safety limit for the relative movement between vehicles.

[0147] As the vehicle speed increases, the driver needs longer to perceive and react to changes ahead, and the braking system also needs longer to slow down the vehicle. Therefore, the faster the vehicle speed, the greater the expected distance between vehicles needs to be to ensure sufficient braking distance in an emergency. Therefore, it is necessary to determine the braking distance based on the vehicle speed and the safe following distance, and increase the braking distance on the basis of the minimum safe distance to ensure sufficient time to react and brake when driving at high speeds.

[0148] The speed difference adjustment item takes into account the effect of the speed difference between the rear and front vehicles on the desired vehicle distance. When the rear vehicle is faster than the front vehicle, in order to avoid rear-end collision, additional vehicle distance is required to cope with possible speed changes. The size of this part of the vehicle distance is proportional to the square root of the speed difference and is affected by vehicle performance (such as maximum acceleration and maximum deceleration).

[0149] It can be seen that the entire formula reflects the dynamic adaptability of the vehicle in the process of following. It dynamically adjusts the expected distance according to real-time changing factors such as vehicle speed, speed difference and vehicle performance to ensure that a sufficient safe distance can be maintained under different driving conditions.

[0150] The expected distance in the car-following model is closely related to the stability of traffic flow. When the vehicle can maintain an appropriate expected distance, the speed fluctuation and density fluctuation in the traffic flow can be reduced, thereby improving the stability and traffic efficiency of the traffic flow. The calculation of the expected distance can often reflect the driver's driving intention and behavior characteristics. By simulating the expected distance under different conditions, the driver's driving behavior can be more accurately reflected, and the predicted moving distance of the sample vehicle can be accurately predicted.

[0151] S12. Determine the distance between the sample vehicle and the preceding vehicle based on the position of the sample vehicle and the position of the preceding vehicle, and determine the vehicle acceleration of the sample vehicle at the first trajectory point based on the expected vehicle spacing and distance between the sample vehicle and the preceding vehicle.

[0152] Specifically, as a feasible implementation method, the vehicle acceleration of the sample vehicle can be determined by the following formula (3):

[0153]

[0154] in, is the vehicle acceleration of the sample vehicle; V0 is the free flow speed; S1 is the distance between the target vehicle and the preceding vehicle.

[0155] It should be understood that acceleration is a physical quantity that describes how fast a vehicle's speed changes. In the car-following model, acceleration directly reflects the vehicle's response to changes in the speed of the vehicle ahead or changes in road conditions. For example, when the vehicle ahead slows down, the vehicle behind needs to slow down accordingly to avoid a collision, and this deceleration process involves changes in acceleration. During vehicle driving, acceleration is not constant and may change with changes in road conditions, traffic conditions, and driver operations.

[0156] S13, determining a predicted moving distance of the sample vehicle based on the vehicle acceleration of the sample vehicle at the first trajectory point, the speed of the sample vehicle, and the time length of the preset period.

[0157] Specifically, as a feasible implementation method, the predicted moving distance of the sample vehicle can be determined by the following formula (4):

[0158]

[0159] in, is the predicted moving distance of the sample vehicle; is the position of the sample vehicle at the first trajectory point; Δt is the length of the preset period.

[0160] Once the vehicle's acceleration and other parameters are known, the predicted distance traveled by the sample vehicle can be determined using basic physical kinematics formulas. These formulas describe the displacement (i.e., distance traveled) of an object at a given acceleration and initial velocity. This process usually involves a comprehensive consideration of vehicle dynamics, driver behavior, and traffic environment.

[0161] It can be seen that the vehicle following model provided in this embodiment predicts and simulates the moving distance of the sample vehicle within a preset time by analyzing the impact of the change in the motion state of the leading vehicle on the driving state of the sample vehicle. This modeling method is closer to the actual driving situation and can more accurately reflect the interaction between vehicles. In addition, it not only considers the basic parameters such as the speed, acceleration and position of the vehicle, but also incorporates multiple factors such as the driver's reaction time, vehicle performance, road conditions, etc., so that the model prediction results are more in line with the actual situation.

[0162] S603: Determine, based on the sample vehicle trajectory, the actual moving distance of the sample vehicle within a preset period of time after the time corresponding to the first trajectory point.

[0163] It should be understood that any model training requires a real value as the expected value so that the output of the model can be closer to the expected value. In the training process of the vehicle following model, the expected value is the actual moving distance of the sample vehicle within a preset period of time after the time corresponding to the first trajectory point.

[0164] S604: Determine a second loss corresponding to the sample vehicle trajectory based on the predicted moving distances and the actual moving distances corresponding to the plurality of first trajectory points.

[0165] After the predicted moving distances and the actual moving distances corresponding to the plurality of first trajectory points in the sample vehicle trajectory are obtained, the second loss corresponding to the sample vehicle trajectory may be determined based on a preset loss function.

[0166] Exemplarily, as a feasible implementation method, the second loss can be determined by the following formula (5).

[0167]

[0168] Among them, L IDE is the second loss corresponding to the sample vehicle trajectory, N is the number of first trajectory points in the sample vehicle trajectory, is the actual moving distance corresponding to the first trajectory point i, is the predicted moving distance corresponding to the first trajectory point i.

[0169] S605 , updating the initial fusion model based on the first loss and the second loss corresponding to the plurality of sample vehicle trajectories until an iteration stop condition is met, thereby obtaining a target data-driven reconstruction model.

[0170] Among them, the initial fusion model includes the initial data-driven reconstruction model and the initial vehicle following model.

[0171] After the first loss and the second loss corresponding to the sample vehicle trajectory, the total loss corresponding to the sample vehicle trajectory can be determined based on the first loss and the second loss, and then the initial fusion model is trained based on the total loss corresponding to multiple sample vehicle trajectories until the iteration stop condition is met, and the target fusion model, that is, the target data-driven reconstruction model and the target vehicle following model, can be obtained.

[0172] As a feasible implementation method, after obtaining the first loss and the second loss corresponding to each sample vehicle trajectory, the total loss corresponding to each sample vehicle trajectory can be obtained by a weighted summation method.

[0173] It can be seen that the method provided in this embodiment uses the vehicle following model to determine the predicted moving distance of the sample vehicle when training the target data-driven reconstruction model, and then determines the second loss corresponding to the sample vehicle trajectory based on the predicted moving distance and the actual moving distance. In this way, during the initial fusion model training process, the initial data-driven reconstruction model can not only mine the vehicle motion features from the sample vehicle trajectory, but also ensure that the prediction results meet the vehicle motion constraints, thereby ensuring the generalization and accuracy of the model.

[0174] In some embodiments, in an intersection scenario, vehicles frequently change lanes, and the running directions of the target vehicle and the preceding vehicle may be inconsistent in the intersection. It is difficult to identify the preceding vehicle of the current vehicle through the grid processing provided by the above embodiment.

[0175] As a possible implementation, see Figure 8 , determine the preceding vehicle information corresponding to the first trajectory point in the sample vehicle trajectory, including:

[0176] S701: Determine a plurality of neighboring vehicles whose distances from a first trajectory point of a sample vehicle are within a preset distance threshold.

[0177] It can be understood that only vehicles whose distance to the sample vehicle is within the preset distance threshold will affect the operation of the sample vehicle, while vehicles outside the preset distance threshold, even if they may be the front vehicle of the sample vehicle in the actual scenario, will not affect the operation of the sample vehicle due to the long distance. Therefore, it is necessary to screen out multiple neighboring vehicles whose distance to the first trajectory point of the sample vehicle is within the preset distance threshold.

[0178] It should be understood that the preset distance threshold is pre-set and can be set according to needs in the actual application process, and the embodiment of the present application does not limit this. Exemplarily, as a feasible implementation method, the preset distance threshold can be 50 meters.

[0179] S702: Determine the running angle of each neighboring vehicle among a plurality of neighboring vehicles and the running angle of the sample vehicle at the first trajectory point.

[0180] It can be understood that the running angle of the vehicle refers to the angle of the vehicle relative to a preset direction, which is used to characterize the deviation of the vehicle's forward direction relative to a reference object.

[0181] S703. Among multiple neighboring vehicles, delete neighboring vehicles whose running angle difference with the sample vehicle is greater than a preset angle difference value to obtain multiple target neighboring vehicles.

[0182] It is understandable that if the running angles of two vehicles differ greatly, it means that the two vehicles are heading in different directions, and vehicles with different heading directions are usually in different lanes, such as a straight-moving vehicle and a left-turning vehicle. Therefore, a portion of vehicles can be selected from multiple neighboring vehicles based on the difference in the running angles of the vehicles, and neighboring vehicles whose running angles differ from the sample vehicle by more than a preset angle difference value can be deleted to obtain multiple target neighboring vehicles.

[0183] It should be understood that the preset angle difference is pre-set and can be set according to needs in the actual application process, and the embodiment of the present application does not limit this. Exemplarily, as a feasible implementation method, the preset angle difference can be 30°.

[0184] As a feasible implementation method, a preliminary screening can be performed based on the running direction of the vehicle to eliminate adjacent vehicles with greatly different running directions. For example, if the running direction of the sample vehicle is north, the vehicles running south, east, and west will be eliminated, and then a secondary screening can be performed based on the running angle of the vehicle to reduce the amount of calculation in the screening step.

[0185] S704. Among multiple target adjacent vehicles, a target adjacent vehicle that is in the same lane as the sample vehicle and is located in front of the sample vehicle is regarded as a preceding vehicle of the sample vehicle.

[0186] It can be understood that after the screening in the above steps, the target adjacent vehicles obtained are all vehicles that are close to the sample vehicle and have a roughly similar running angle, but some of these vehicles may be in a different lane from the sample vehicle or behind the sample vehicle. Therefore, after obtaining multiple target adjacent vehicles, the target adjacent vehicle that is in the same lane as the sample vehicle and is located in front of the sample vehicle is taken as the preceding vehicle of the sample vehicle.

[0187] S705 , obtaining the position and speed of the preceding vehicle of the sample vehicle, and obtaining preceding vehicle information corresponding to the first trajectory point.

[0188] That is, the position and speed of the preceding vehicle of the sample vehicle are used as the preceding vehicle information corresponding to the first trajectory point.

[0189] It can be seen that the method provided in this embodiment first excludes vehicles that are far away from the sample vehicle through distance screening, thus avoiding the analysis of a large number of vehicles that are irrelevant to the current situation, thereby greatly reducing the amount of calculation and improving the processing speed of the entire process. Further, through the screening of the running angle, vehicles whose driving direction or trajectory is greatly different from that of the sample vehicle are excluded. Even if these vehicles are close, they are unlikely to become the front vehicle of the sample vehicle. This step further narrows the range of potential front vehicles and improves the accuracy of subsequent judgments. After narrowing the range, focus on identifying the target adjacent vehicle that is in the same lane as the sample vehicle and in front of it. This dual judgment based on lane and position can very accurately identify the front vehicle of the sample vehicle, thereby ensuring that the acquired front vehicle information is accurate.

[0190] In other words, the information about the vehicle ahead provided by this method is highly accurate, which helps the vehicle make more reasonable and safe driving decisions and improves overall driving safety. This method is not only applicable to structured roads such as highways, but can also operate effectively in scenarios such as urban roads and complex traffic environments. By flexibly adjusting the thresholds of distance and operating angle, it can meet the needs of different scenarios and has strong adaptability and flexibility.

[0191] The above mainly introduces the solution provided by the embodiment of the present application from the perspective of the method. In order to achieve the above functions, it includes hardware structures and / or software modules corresponding to the execution of each function. It should be easy to realize that the technical goals in this field are combined with the units and algorithm steps of each example described in the embodiments disclosed in this article, and the present application can be implemented in the form of hardware or a combination of hardware and computer software. Whether a function is executed in the form of hardware or computer software driving hardware depends on the specific application and design constraints of the technical solution. Professional and technical goals can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.

[0192] The present application also provides a vehicle trajectory reconstruction device, such as Fig. 9 As shown, the vehicle trajectory reconstruction device 80 includes: an acquisition module 81 , an input module 82 and a determination module 83 .

[0193] Among them, the acquisition module 81 is used to obtain the initial vehicle trajectory of the target vehicle, the road network data and the signal light state data of the area where the target vehicle is located; the input module 82 is used to input the initial vehicle trajectory, the road network data and the signal light state data into the target data-driven reconstruction model, and obtain multiple target trajectory points corresponding to the initial vehicle trajectory through the target data-driven reconstruction model; the determination module 83 is used to determine the reconstructed vehicle trajectory corresponding to the initial vehicle trajectory based on the initial vehicle trajectory and the multiple target trajectory points.

[0194] The target data-driven reconstruction model obtains multiple target trajectory points through the following steps: extracting vehicle motion features from the initial vehicle trajectory; fusing the signal light state data with the vehicle motion features, and predicting multiple trajectory points of the target vehicle based on the fused data; constraining the coordinates of multiple trajectory points based on the road network data, and obtaining multiple target trajectory points that conform to the road network data.

[0195] In some embodiments, the acquisition module 81 is specifically used to obtain the original vehicle trajectory of the target vehicle; among multiple trajectory points of the original vehicle trajectory, identify abnormal trajectory points whose trajectory points do not conform to the road range to which the trajectory points belong; delete the abnormal trajectory points in the original vehicle trajectory to obtain the initial vehicle trajectory.

[0196] In some embodiments, the vehicle trajectory reconstruction device 80 also includes: a division module, used to divide the initial vehicle trajectory into multiple trajectory segments; an identification module, used to identify whether there are abnormal trajectory segments in the multiple trajectory segments; the abnormal trajectory segments have one or more of the following characteristics: the existence of a corresponding position point with a residence time higher than a preset threshold, multiple trajectory points with the same position, trajectory jitter or trajectory offset; wherein the trajectory jitter is used to indicate that the running direction angle of the target vehicle changes multiple times within a preset time period, and the trajectory offset is used to indicate that the distance between multiple trajectory points in the abnormal trajectory segment and the center line of the lane to which the trajectory points belong exceeds a preset distance; a deletion module, used to delete the abnormal trajectory segments in the initial vehicle trajectory.

[0197] In some embodiments, the target data driven reconstruction model includes an encoder and a decoder, the encoder is used to extract vehicle motion features in the initial vehicle trajectory; the decoder includes a trajectory point prediction layer and a predicted position stack, the trajectory point prediction layer is used to fuse signal light state data with vehicle motion features, and predict multiple trajectory points of the target vehicle based on the fused data; the predicted position stack is used to constrain the coordinates of multiple trajectory points based on road network data to obtain multiple target trajectory points that conform to the road network data.

[0198] In some embodiments, the initial vehicle trajectory includes multiple trajectory points, each trajectory point corresponds to a trajectory point data, and the trajectory point data includes: collection time, longitude and latitude position, vehicle speed, running angle and lane of the vehicle; road network data includes intersections, sections, lanes and topological relationships on the road; signal light state data includes green light duration, red light duration and green light utilization rate corresponding to the lane where the target vehicle is located.

[0199] The embodiment of the present application also provides a data-driven reconstruction model training device, including: an acquisition module, a removal module, an input module, a determination module, and an update module. It should be noted that the data-driven reconstruction model training device and the vehicle trajectory reconstruction device in the above embodiment can be the same device or different devices, and the embodiment of the present application does not limit this.

[0200] Among them, the acquisition module is used to obtain multiple training samples, each of which includes: a sample vehicle trajectory, road network data and signal light state data corresponding to the sample vehicle trajectory. The removal module is used to remove some trajectory points in the sample vehicle trajectory. The input module is used to input the sample vehicle trajectory after the trajectory points are removed, the road network data and signal light state data corresponding to the sample vehicle trajectory into the initial data-driven reconstruction model to obtain multiple target trajectory points corresponding to the sample vehicle trajectory. The determination module is used to calculate the first loss corresponding to the sample vehicle trajectory based on the partial trajectory points removed from the sample vehicle trajectory and the multiple target trajectory points corresponding to the sample vehicle trajectory. The update module is used to update the initial data-driven reconstruction model through the first losses corresponding to the multiple sample vehicle trajectories until the iteration stop condition is met to obtain the target data-driven reconstruction model.

[0201] In some embodiments, the update module is specifically used to determine the position, speed and leading vehicle information of the sample vehicle corresponding to the first trajectory point in the sample vehicle trajectory; the first trajectory point is any trajectory point except the last trajectory point in the sample vehicle trajectory; the leading vehicle information includes the position and speed of the leading vehicle of the sample vehicle; the leading vehicle of the sample vehicle is a vehicle that is in the same lane as the sample vehicle and is in front of the sample vehicle; the position, speed and leading vehicle information of the sample vehicle corresponding to the first trajectory point are input into the initial vehicle following model, and the initial vehicle following model is used to predict the predicted moving distance of the sample vehicle within a preset time period after the time corresponding to the first trajectory point; based on the sample vehicle trajectory, the actual moving distance of the sample vehicle within the preset time period after the time corresponding to the first trajectory point is determined; based on the predicted moving distance and the actual moving distance corresponding to multiple first trajectory points, the second loss corresponding to the sample vehicle trajectory is determined; based on the first loss and the second loss corresponding to the multiple sample vehicle trajectories, the initial fusion model is updated until the iteration stop condition is met to obtain the target data-driven reconstruction model, and the initial fusion model includes the initial data-driven reconstruction model and the initial vehicle following model.

[0202] In some embodiments, the update module is specifically used to determine multiple neighboring vehicles whose distances from the first trajectory point of the sample vehicle are within a preset distance threshold; determine the running angle of each neighboring vehicle among the multiple neighboring vehicles, and the running angle of the sample vehicle at the first trajectory point; among the multiple neighboring vehicles, delete the neighboring vehicles whose running angle difference with the sample vehicle is greater than the preset angle difference, and obtain multiple target neighboring vehicles; among the multiple target neighboring vehicles, take the target neighboring vehicle that is in the same lane as the sample vehicle and is located in front of the sample vehicle as the preceding vehicle of the sample vehicle; obtain the position and speed of the preceding vehicle of the sample vehicle, and obtain the preceding vehicle information corresponding to the first trajectory point.

[0203] In some embodiments, the initial vehicle following model predicts the predicted moving distance of the sample vehicle in the following manner: determining the expected vehicle distance between the sample vehicle and the preceding vehicle based on the speed of the sample vehicle and the speed of the preceding vehicle; determining the distance between the sample vehicle and the preceding vehicle based on the position of the sample vehicle and the position of the preceding vehicle, and determining the vehicle acceleration of the sample vehicle at the first trajectory point based on the expected vehicle distance and the distance between the sample vehicle and the preceding vehicle; determining the predicted moving distance of the sample vehicle based on the vehicle acceleration of the sample vehicle at the first trajectory point, the speed of the sample vehicle, and the length of a preset time period; the expected vehicle distance is calculated by the following formula:

[0204]

[0205] Among them, S * is the expected vehicle distance; s0 is the minimum safe distance; T is the safe following distance; a is the maximum acceleration of the vehicle; b is the maximum deceleration of the vehicle; is the speed of the sample vehicle at the first trajectory point; is the speed of the preceding vehicle.

[0206] In some embodiments, the update module is specifically used to determine multiple neighboring vehicles whose distances from the first trajectory point of the sample vehicle are within a preset distance threshold; determine the running angle of each neighboring vehicle among the multiple neighboring vehicles, and the running angle of the sample vehicle at the first trajectory point; among the multiple neighboring vehicles, delete the neighboring vehicles whose running angle difference with the sample vehicle is greater than the preset angle difference, and obtain multiple target neighboring vehicles; among the multiple target neighboring vehicles, take the target neighboring vehicle that is in the same lane as the sample vehicle and is located in front of the sample vehicle as the preceding vehicle of the sample vehicle; obtain the position and speed of the preceding vehicle of the sample vehicle, and obtain the preceding vehicle information corresponding to the first trajectory point.

[0207] The present application also provides an electronic device. Fig.10The electronic device 201 includes: one or more memories 121, one or more processors 122, a communication bus 123 and a communication interface 124. The processor 122 is connected to the memory 121 via the bus 123; the one or more memories 121 are used to store computer program codes, and the computer program codes include computer instructions; when the one or more processors 122 execute the computer instructions, the electronic device 201 executes the vehicle trajectory reconstruction method provided by the above embodiment.

[0208] Optionally, the memory 121 may be a non-temporary computer-readable storage medium, for example, the non-temporary computer-readable storage medium may be a read-only memory (ROM), a random access memory (RAM), a CD-ROM, a tape, a floppy disk, and an optical data storage device, etc. The embodiments of the present application do not impose any restrictions on this.

[0209] The processor 122 can be a central processing unit (CPU), a general-purpose processor network processor (NP), a digital signal processor (DSP), a microprocessor, a microcontroller, a programmable logic device (PLD) or any combination thereof, and the embodiments of the present application do not impose any restrictions on this.

[0210] The communication bus 123 may be an industry standard architecture (ISA) bus, a peripheral component interconnect (PCI) bus, or an extended industry standard architecture (EISA) bus. The communication bus 123 may be divided into an address bus, a data bus, a control bus, etc. For ease of representation, Fig.10 Just one thick line is used, but it does not mean that there is only one bus or one type of communication bus.

[0211] The communication interface 124 uses any transceiver or other device for communicating with other devices or communication networks, such as control systems, radio access networks (RAN), wireless local area networks (WLAN), etc.

[0212] The embodiments of the present application also provide a computer program product including one or more instructions, where the one or more instructions are stored in a memory of a computer device and executed by a processor to complete the various processes of the above embodiments.

[0213] An embodiment of the present application further provides a computer-readable storage medium, which includes computer-executable instructions. When the computer-executable instructions are executed on a computer, the computer executes the vehicle trajectory reconstruction method provided in the above embodiment.

[0214] Through the description of the above implementation methods, technical personnel in the relevant field can clearly understand that for the convenience and simplicity of description, only the division of the above-mentioned functional modules is used as an example. In actual applications, the above-mentioned functions can be assigned to different functional modules as needed, that is, the internal structure of the device can be divided into different functional modules to complete all or part of the functions described above.

[0215] In the several embodiments provided in the present application, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic, for example, the division of modules or units is only a logical function division, and there may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another device, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0216] The units described as separate components may or may not be physically separated, and the components shown as units may be one physical unit or multiple physical units, that is, they may be located in one place or distributed in multiple different places. Some or all of the units may be selected according to actual needs to achieve the purpose of the present embodiment.

[0217] In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0218] If 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 readable storage medium. Based on this understanding, the technical solution of the embodiment of the present application is essentially or the part that contributes to the prior art or all or part of the technical solution can be embodied in the form of a software product, which is stored in a storage medium and includes several instructions for a device or processor to execute all or part of the steps of each embodiment method of the present application. The aforementioned storage medium includes: various media that can store program codes, such as USB flash drives, mobile hard drives, ROM, RAM, magnetic disks or optical disks.

[0219] The above are only specific implementations of the present application, but the protection scope of the present application is not limited thereto, and any changes or substitutions within the technical scope disclosed in the present application should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

Claims

1. A vehicle trajectory reconstruction method, characterized in that: include: Obtaining the initial vehicle trajectory of the target vehicle, the road network data and the signal light state data of the area where the target vehicle is located; Inputting the initial vehicle trajectory, the road network data and the signal light state data into a target data driven reconstruction model, and obtaining a plurality of target trajectory points corresponding to the initial vehicle trajectory through the target data driven reconstruction model; Determining a reconstructed vehicle trajectory corresponding to the initial vehicle trajectory based on the initial vehicle trajectory and the plurality of target trajectory points; The target data driven reconstruction model obtains the multiple target trajectory points through the following steps: extracting vehicle motion features from the initial vehicle trajectory; Fusing the signal light state data with the vehicle motion characteristics, and predicting multiple trajectory points of the target vehicle based on the fused data; The coordinates of the multiple trajectory points are constrained based on the road network data to obtain multiple target trajectory points that conform to the road network data.

2. The method according to claim 1, characterized in that The obtaining of the initial vehicle trajectory of the target vehicle comprises: Obtaining an original vehicle trajectory of the target vehicle; Among the plurality of trajectory points of the original vehicle trajectory, identifying an abnormal trajectory point that does not conform to a road range to which the trajectory point belongs; The abnormal trajectory points in the original vehicle trajectory are deleted to obtain an initial vehicle trajectory.

3. The method according to claim 1 or 2, characterized in that: Before inputting the initial vehicle trajectory, the road network data and the signal light state data into the target data driven reconstruction model, the method further includes: dividing the initial vehicle trajectory into a plurality of trajectory segments; Identify whether there is an abnormal trajectory segment among the multiple trajectory segments; the abnormal trajectory segment has one or more of the following characteristics: the presence of a position point whose corresponding stay time is higher than a preset threshold, multiple trajectory points with the same position, trajectory jitter or trajectory offset; wherein the trajectory jitter is used to indicate that the running direction angle of the target vehicle changes multiple times within a preset time period, and the trajectory offset is used to indicate that the distance between the multiple trajectory points in the abnormal trajectory segment and the center line of the lane to which the trajectory points belong exceeds a preset distance; Abnormal trajectory segments in the initial vehicle trajectory are deleted.

4. The method according to claim 1, characterized in that: The target data driven reconstruction model includes an encoder and a decoder, wherein the encoder is used to extract vehicle motion features in the initial vehicle trajectory; The decoder includes a trajectory point prediction layer and a prediction position stack, wherein the trajectory point prediction layer is used to fuse the signal light state data with the vehicle motion characteristics, and predict multiple trajectory points of the target vehicle based on the fused data; The predicted position stack is used to constrain the coordinates of the multiple trajectory points based on the road network data to obtain multiple target trajectory points that conform to the road network data.

5. The method according to claim 1, characterized in that The target data driven reconstruction model is trained in the following way: Acquire multiple training samples, wherein each training sample includes: a sample vehicle trajectory, road network data and signal light state data corresponding to the sample vehicle trajectory; Removing some track points from the sample vehicle track; Inputting the sample vehicle trajectory after removing the trajectory points, the road network data corresponding to the sample vehicle trajectory, and the signal light state data into the initial data-driven reconstruction model to obtain a plurality of target trajectory points corresponding to the sample vehicle trajectory; Calculating a first loss corresponding to the sample vehicle trajectory based on a portion of trajectory points removed from the sample vehicle trajectory and a plurality of target trajectory points corresponding to the sample vehicle trajectory; The initial data-driven reconstruction model is updated by using the first losses corresponding to the plurality of sample vehicle trajectories until an iteration stop condition is met, thereby obtaining the target data-driven reconstruction model.

6. The method according to claim 5, characterized in that The updating of the initial data-driven reconstruction model by first losses corresponding to the plurality of sample vehicle trajectories until an iteration stop condition is satisfied to obtain the target data-driven reconstruction model comprises: Determine the position, speed and preceding vehicle information of the sample vehicle corresponding to the first trajectory point in the sample vehicle trajectory; the first trajectory point is any trajectory point except the last trajectory point in the sample vehicle trajectory; the preceding vehicle information includes the position and speed of the preceding vehicle of the sample vehicle; the preceding vehicle of the sample vehicle is a vehicle in the same lane as the sample vehicle and in front of the sample vehicle; Inputting the position, speed and preceding vehicle information of the sample vehicle corresponding to the first trajectory point into an initial vehicle following model, and predicting the predicted moving distance of the sample vehicle within a preset period after the time corresponding to the first trajectory point by the initial vehicle following model; Determine, based on the sample vehicle trajectory, an actual moving distance of the sample vehicle within a preset time after the time corresponding to the first trajectory point; determining a second loss corresponding to the sample vehicle trajectory based on the predicted moving distances and the actual moving distances corresponding to the plurality of first trajectory points; The initial fusion model is updated based on the first loss and the second loss corresponding to the plurality of sample vehicle trajectories until the iteration stop condition is met to obtain the target data-driven reconstruction model, wherein the initial fusion model includes the initial data-driven reconstruction model and the initial vehicle following model.

7. The method according to claim 6, characterized in that The initial vehicle following model predicts the predicted moving distance of the sample vehicle in the following manner: Determining an expected vehicle distance between the sample vehicle and the preceding vehicle according to the speed of the sample vehicle and the speed of the preceding vehicle; Determining the distance between the sample vehicle and the preceding vehicle based on the position of the sample vehicle and the position of the preceding vehicle, and determining the vehicle acceleration of the sample vehicle at the first trajectory point based on the expected vehicle spacing and the distance between the sample vehicle and the preceding vehicle; determining a predicted moving distance of the sample vehicle based on the vehicle acceleration of the sample vehicle at the first trajectory point, the speed of the sample vehicle, and the time length of the preset period; The expected vehicle distance is calculated by the following formula: Among them, S * is the desired vehicle distance; s0 is the minimum safe distance; T is the safe following distance; a is the maximum acceleration of the vehicle; b is the maximum deceleration of the vehicle; is the speed of the sample vehicle at the first trajectory point; is the speed of the preceding vehicle.

8. The method according to claim 6, characterized in that Determining the preceding vehicle information corresponding to the first trajectory point in the sample vehicle trajectory includes: determining a plurality of neighboring vehicles whose distances from the first trajectory point of the sample vehicle are within a preset distance threshold; Determine the running angle of each of the plurality of neighboring vehicles and the running angle of the sample vehicle at the first trajectory point; Among the plurality of adjacent vehicles, adjacent vehicles whose running angle difference with the sample vehicle is greater than a preset angle difference are deleted to obtain a plurality of target adjacent vehicles; Among the plurality of target adjacent vehicles, a target adjacent vehicle that is in the same lane as the sample vehicle and is located in front of the sample vehicle is taken as a preceding vehicle of the sample vehicle; The position and speed of the preceding vehicle of the sample vehicle are obtained to obtain the preceding vehicle information corresponding to the first trajectory point.

9. The method according to claim 1, characterized in that: The initial vehicle trajectory includes a plurality of trajectory points, each of which corresponds to a trajectory point data, and the trajectory point data includes: acquisition time, longitude and latitude position, vehicle speed, running angle and lane where the vehicle is located; The road network data includes intersections, road sections, lanes and topological relationships on the road; The signal light state data includes the green light duration, red light duration and green light utilization rate corresponding to the lane where the target vehicle is located.

10. An electronic device, characterized in that: include: one or more processors; one or more memories; Wherein, the one or more memories are used to store computer program codes, and the computer program codes include computer instructions. When the one or more processors execute the computer instructions, the electronic device executes the vehicle trajectory reconstruction method as described in any one of claims 1 to 9.

Images