A method and device for predicting intersection trajectory based on strategic intention

Abstract

The present invention discloses a method for predicting intersection trajectories based on strategic intentions, comprising the following steps: Step 1, setting up road-side equipment at intersections, and collecting historical trajectory information of traffic participants including vehicles and road environments, obtaining high-precision maps of the area, and obtaining raw data; Step 2, vectorizing the scene to obtain global interactive features; Step 3, constructing an input matrix according to the divided areas; Step 4, decoding through a decoder to obtain multiple possible trajectories, and selecting the trajectory that best meets the driving intention as the final predicted trajectory; Step 5, training with data collected at road intersections as samples; Step 6, performing trajectory prediction on the road-side equipment. The present invention does not require driving intention estimation, but uses the determined driving intention at the intersection to select a predicted trajectory that meets the driver's expectations, avoiding the uncertainty caused by individual differences, and greatly improving the efficiency and accuracy of trajectory prediction.

Description

Technical Field

[0001] The present invention relates to a trajectory prediction method, and in particular to an intersection trajectory prediction method and device based on strategic intention. Background Art

[0002] Trajectory prediction can provide more decision-making basis for drivers and vehicle warning systems, and is of great significance for driving safety assessment and vehicle path planning. In addition, in traffic flow, vehicle trajectory prediction provides a good idea for solving practical problems such as road congestion and obstacle avoidance for traffic participants.

[0003] The challenges faced by vehicle trajectory prediction are, on the one hand, the high complexity of road intersections, and on the other hand, the great uncertainty caused by individual differences in drivers. Therefore, properly considering these two challenges can provide more accurate and reliable prediction trajectories for intelligent vehicles.

[0004] Although the existing technology usually takes into account the interaction of surrounding vehicles, it still lacks globality. For example, the patent CN114005280 A relies on vehicle-mounted equipment to obtain information about surrounding vehicles of the candidate vehicle, which can only cover a local area. In addition, the current data-driven prediction method is mostly used for simple scenarios, which may not be applicable to complex and changeable scenarios.

[0005] In reality, even if the vehicles have similar historical trajectories, the driver's driving intention may lead to different future trajectories, and the predicted trajectories may diverge. There are some methods, such as CN112347567 B, which first identifies the driving intention and then predicts the trajectory based on the driving intention. However, this type of method is prone to interference with trajectory prediction and lacks practicality, because multiple different results may be generated within a continuous driving intention identification cycle. Academically, driving intention is classified into strategic intention, tactical intention, and operational intention. The use of strategic intention is often ignored by researchers. If strategic intention is used properly, the expectation of stable driving behavior can be obtained, which is more beneficial and auxiliary trajectory prediction. Therefore, how to use traffic rules and strategic intentions in complex traffic scenarios to increase the certainty and accuracy of vehicle trajectory prediction is a technical problem that needs to be solved urgently. Summary of the invention

[0006] The purpose of the present invention is to increase the certainty of trajectory prediction at complex intersections, thereby improving the trajectory prediction accuracy and further providing more decision-making basis for the driver and the vehicle's danger warning system.

[0007] The technical solution of the present invention is to provide a method for predicting intersection trajectories based on strategic intent, comprising the following steps:

[0008] Step 1: Set up roadside equipment at the intersection, collect historical trajectory information of traffic participants including vehicles and road environment, obtain high-precision maps of the area, and obtain original data;

[0009] Step 2: vectorize the scene, encode each constructed sub-graph, generate a global graph, and obtain global interaction features;

[0010] Step 3: construct an input matrix according to the divided areas, and invalidate the restricted areas according to traffic rules;

[0011] Step 4: Obtain multiple possible trajectories through decoding by the decoder, and select the trajectory that best meets the driving intention as the final predicted trajectory;

[0012] Step 5: In the training phase, the data collected at the road intersections are used as samples for training;

[0013] Step 6: Perform trajectory prediction on the road-side device and transmit the results to each vehicle.

[0014] Further, step 1 is implemented by:

[0015] Step 1.1, perform panoramic acquisition of the intersection with a sampling frequency of 10 Hz;

[0016] Step 1.2, obtain the high-precision map of the intersection, and coordinate the scene information collected in step 1.1 on the high-precision map;

[0017] Step 1.3, the road-side device receives in advance the strategic intention sent by the vehicle through the sending module, thereby obtaining its traveling direction at the intersection, that is, the driving intention of the intersection.

[0018] Furthermore, step 2 is implemented by:

[0019] Step 2.1, divide the intersection into five areas, which are numbered k, k = 1, 2, 3, 4, 5;

[0020] Step 2.2: Vectorize the information in each area, abstract each scene information such as vehicle trajectory, road, and lane line into a polyline, and convert the vector V that composes the polyline into i The head and tail coordinates, semantic information, attribute number and region number of are represented as a two-dimensional matrix:

[0021]

[0022] Among them, V i The first column represents the starting point coordinates;

[0023] The second column represents the end point coordinates;

[0024] The third column represents the attribute and sampling frequency, i.e., the semantic label;

[0025] The fourth column represents the same attribute number;

[0026] The fifth column represents the region number;

[0027] Step 2.3: V with the same attribute and the same i i Connect to form a line subgraph

[0028] Step 2.4: Encode the line subgraph features. The encoding method is as follows:

[0029]

[0030] in,

[0031] It represents the use of one-dimensional convolution to encode the input features;

[0032] and Represents maximum pooling and mean pooling respectively;

[0033] is a linear mapping.

[0034] Furthermore, step 3 is implemented by:

[0035] Step 3.1: Use the line subgraph features encoded in step 2.4 to construct an input matrix according to the p subgraphs contained in the region k where it is located:

[0036]

[0037] Step 3.2: According to the no-travel restrictions obtained by the roadside equipment, the input matrix of the corresponding area is invalidated. The processing method is as follows:

[0038]

[0039] in, Represents the feature input of the entire intersection, which is composed of the above five input matrices, where 0 is the non-road area;

[0040] Θ is the prohibited area identifier for each area, which is 0 when prohibited and 1 otherwise;

[0041] If the area is forbidden when k=2, the failure processing is performed according to formula (6) to obtain the final input matrix

[0042]

[0043] Step 3.3: The subgraph features in are taken as nodes in the global interaction graph GNN:

[0044]

[0045] Among them, GNN(·) is a graph neural network, which is implemented through the self-attention mechanism;

[0046] is the adjacency matrix, which represents the spatial distance between nodes;

[0047] To extract the global interaction features, Divided into historical input features and future real features

[0048] Furthermore, step 4 is implemented by:

[0049] Step 4.1: From the probability distribution P that conforms to the Gaussian distribution z Sample the latent space variable z in i , and through a linear layer After matching the dimensions with Splicing As shown in formula (7), After decoding by the decoder, a trajectory s is obtained i ;

[0050]

[0051] Step 4.2: Repeat step 4.1 N times for each target vehicle to obtain a set of possible future trajectories. is the decoder, T pred represents the time step of the predicted trajectory;

[0052] Step 4.3: Determine the lane that the target vehicle will pass through based on its driving intention, and use the endpoint of the lane centerline as the reference point for screening future trajectories; take 6 coordinate points at equal intervals for each decoded future trajectory, and calculate the Euclidean distance between each point and the reference point, sum the results, and the trajectory corresponding to the minimum sum is the final predicted trajectory.

[0053] Furthermore, step 5 is implemented by:

[0054] Step 5.1: and After concatenation, input an MLP layer;

[0055] Step 5.2: The result of step 5.1 is estimated by conditional variational autoencoder (CVAE) to be and the variance is The latent variable z i , is a Gaussian distribution; z i and The reconstructed future trajectory is obtained as the input of the decoder LSTM network

[0056] Furthermore, step 6 is implemented by:

[0057] The trajectory of each target vehicle at the intersection is predicted on the road-side device, and all predicted trajectories are sent to each target vehicle by the communication device. The predicted trajectory visualization results can be further sent to the vehicle display.

[0058] The beneficial effects of the present invention are:

[0059] (1) The present invention collects and processes the entire intersection information through road-side equipment. Compared with the local information that can be obtained by vehicle-mounted sensor equipment, it better models the interaction between traffic participants and improves the accuracy of trajectory prediction.

[0060] (2) The data collection, storage, trajectory prediction, etc. in the present invention are all completed by the road-side equipment. The trajectory prediction method is implemented through Example 2, so there is no need for the vehicle side to collect and process data, saving computing power. In addition, the prediction results are sent to the vehicle-side receiving module through the vehicle-road collaborative method, and the received results can be further visualized through the display, which increases the driver's trust in the trajectory prediction system and improves driving safety.

[0061] (3) The present invention uses traffic rules to regulate input data. The roadside equipment selectively invalidates the data of the prohibited area based on the acquired traffic restriction information, shields the line subgraphs that do not contribute to the result when constructing the input matrix in a hierarchical manner, and can be used as a basis for screening candidate prediction trajectories.

[0062] (4) The present invention learns the vehicle's intention at the intersection by sending the vehicle's strategic intention to the roadside device in advance. This method does not require driving intention estimation, but uses the exact vehicle direction to select a predicted trajectory that meets the driver's expectations, avoiding the uncertainty caused by individual differences and greatly improving the efficiency and accuracy of trajectory prediction. BRIEF DESCRIPTION OF THE DRAWINGS

[0063] Figure 1 It is a flow chart of the present invention;

[0064] Figure 2 It is a schematic diagram of vehicle-roadside equipment for complex intersection trajectory prediction based on strategic intent;

[0065] Figure 3 It is a schematic diagram of the intersection division area;

[0066] Figure 4 It is a schematic diagram of the scene information vectorization method;

[0067] Figure 5 It is the structure diagram of the line subgraph feature encoder;

[0068] Figure 6 It is a schematic diagram of the final prediction trajectory selection; DETAILED DESCRIPTION

[0069] In order to more clearly understand the above-mentioned purposes, features and advantages of the present application, the present application is further described in detail below in conjunction with the accompanying drawings and specific implementation methods. It should be noted that the embodiments of the present application and the features in the embodiments can be combined with each other without conflict.

[0070] In the following description, many specific details are elaborated to facilitate a full understanding of the present application. However, the present application can also be implemented using a driver different from the driver described herein. Therefore, the scope of protection of the present application is not limited to the specific embodiments disclosed below.

[0071] like Figure 1 As shown, this embodiment 1 provides a vehicle-road end device for intersection trajectory prediction based on strategic intent, and each device is only responsible for the intersection;

[0072] The road-side equipment consists of a data acquisition module, a processor, a communication device, a memory and a server; the road-side equipment is set up at an intersection; the data acquisition module is responsible for collecting intersection information, and inputting the information and the high-precision map of the intersection into the storage device for storage; in addition to storing relevant intersection information, the memory is also responsible for storing the trajectory prediction algorithm program; the server is responsible for training and reasoning;

[0073] The vehicle-side device includes a receiving module, a sending module and a display;

[0074] The receiving module is responsible for receiving the trajectory prediction results of the road section equipment;

[0075] The sending module is responsible for sending the driver's strategic intention, which is the driver's desired destination, and is used to pre-judge his driving intention (left turn, right turn or straight ahead) when passing through the intersection;

[0076] The display is responsible for visualizing the received predicted trajectories.

[0077] Example 2 provides an intersection trajectory prediction method based on strategic intent, which can serve human-machine co-driving vehicles and higher-level autonomous driving vehicles.

[0078] The core idea of ​​the method described in this embodiment 2 is to increase the certainty in the trajectory prediction problem, because the driver has a strong subjectivity, and the driving activities with the driver's participation will have a large uncertainty. Increasing certainty is reflected in two aspects, one is to use the driver's intention, and the other is to use traffic rules.

[0079] Regarding the use of driving intention, this embodiment 2 does not estimate the driving intention in the way of the prior art, but directly obtains the accurate driving intention at the intersection. Specifically, the driving intention is obtained in the following way:

[0080] Assuming that we know that a driver is driving from A to B, that is, the strategic intention, then when passing a certain intersection, whether the driver turns left, turns right, or goes straight can be considered to be certain. This specific intersection is the intersection mentioned in the patent, and the determined direction of travel is the driving intention at the intersection. To obtain the driver's strategic intention, you can read the vehicle navigation information, or ask the driver's destination through the smart cockpit voice assistant.

[0081] For the use of traffic rules, such as traffic lights, since the direction of the predicted candidate trajectories is divergent, traffic restrictions can be used to filter out some of the candidate trajectories. In addition, the input data of the trajectory prediction algorithm can be preliminarily processed according to traffic restrictions. Since the input algorithm is the data of traffic participants at the entire intersection, rather than the data of the local area obtained by the vehicle-mounted sensors, the data of the entire intersection is not all valid. If one of the intersections (for example, an intersection with 4 intersections) is restricted by a red light, then the data of this intersection will not contribute to the trajectory prediction, but may interfere with it. Therefore, the data of this intersection needs to be invalidated in the input algorithm model, which can be dynamically adjusted according to the changes of traffic signals, etc.

[0082] It should be noted that data acquisition and trajectory prediction are all completed by the road-side device described in the embodiment of the present invention, rather than being calculated by the device on the vehicle as in the conventional method. In the embodiment of the present invention, the vehicle and the road-side device communicate, the vehicle sends the strategic intention to the road-side device, and the road-side device sends the predicted results to the vehicle. The display on the vehicle is used to visualize the prediction results received by the vehicle, which is convenient for the driver to view in real time and improve the driver's understanding of the driving system. Optionally, the intelligent driving system on the vehicle can also use the received final trajectory prediction results for pre-planning, early warning systems, etc.

[0083] In this embodiment, a method for predicting intersection trajectories based on strategic intent includes the following steps:

[0084] Step 1: Set up roadside equipment at the intersection, collect historical trajectory information of traffic participants including vehicles and road environment, obtain high-precision maps of the area, and obtain original data;

[0085] Further, step 1 is implemented by:

[0086] Step 1.1: Perform panoramic capture of the intersection with a sampling frequency of 10 Hz. During the training phase, each scene lasts for 5 seconds, with the first 2 seconds used as the historical trajectory and the last 3 seconds used as the trajectory prediction part of the algorithm; during the inference phase, 2 seconds are used as input and 3 seconds of prediction results are output;

[0087] The intersection is captured in full view, including the spatial and temporal distribution of traffic participants, road traffic facilities and markings in the area. The above content constitutes the scene information of the collection period, and the sampling frequency is 10Hz. In the training phase, each scene lasts for 5 seconds, with the first 2 seconds as the historical trajectory and the last 3 seconds as the trajectory prediction part of the algorithm; in the inference phase, 2 seconds are used as input and 3 seconds of prediction results are output;

[0088] Step 1.2, obtain the high-precision map of the intersection, and coordinate the scene information collected in step 1.1 on the high-precision map;

[0089] In this step, a high-precision map of the intersection is obtained, and the scene information collected in step 1.1 is matched with the high-precision map frame by frame through coordinate conversion, clock synchronization, and trajectory combination, so that the coordinates of the traffic participants are calibrated on the high-precision map using relative positions.

[0090] Step 1.3, the road-side device receives in advance the strategic intention sent by the vehicle through the sending module, thereby obtaining its traveling direction at the intersection, that is, the driving intention of the intersection.

[0091] Step 2: vectorize the scene, encode each constructed sub-graph, generate a global graph, and obtain global interaction features;

[0092] Furthermore, step 2 is implemented by:

[0093] Step 2.1: Figure 3 As shown, the intersection is divided into five areas, which are numbered k, k = 1, 2, 3, 4, 5;

[0094] Step 2.2: Figure 4 As shown in the figure, the information in each area is vectorized, and each scene information such as vehicle trajectory, road, lane line, etc. is abstracted into a polyline, and the vector V that constitutes the polyline is i The head and tail coordinates, semantic information, attribute number and region number of are represented as a two-dimensional matrix:

[0095]

[0096] Among them, V i The first column represents the starting point coordinates;

[0097] The second column represents the end point coordinates;

[0098] The third column represents the attribute and sampling frequency (determines the number of vectors m that make up the polyline), i.e., the semantic label;

[0099] The fourth column represents the number of the same attribute (for example, the left and right lane lines of the lane are numbered 1 and 2);

[0100] The fifth column represents the region number;

[0101] Step 2.3: V with the same attribute and the same i i Connect to form a line subgraph

[0102] Step 2.4: Encode the line subgraph features. The encoding method is as follows:

[0103]

[0104] in,

[0105] It represents the use of one-dimensional convolution to encode the input features;

[0106] and Represents maximum pooling and mean pooling respectively;

[0107] is a linear mapping;

[0108] The encoder structure used is as follows Figure 5 shown.

[0109] Step 3: construct an input matrix according to the divided areas, and perform invalidation processing on the restricted areas according to traffic rules (such as traffic lights, etc.);

[0110] Furthermore, step 3 is implemented by:

[0111] Step 3.1: Use the line subgraph features encoded in step 2.4 to construct an input matrix according to the p subgraphs contained in the region k where it is located:

[0112]

[0113] Step 3.2: According to the no-travel restrictions obtained by the roadside equipment, the input matrix of the corresponding area is invalidated. The processing method is as follows:

[0114]

[0115] in, Represents the feature input of the entire intersection, which is composed of the above five input matrices, where 0 is the non-road area;

[0116] Θ is the prohibited area identifier for each area, which is 0 when prohibited and 1 otherwise;

[0117] If the area is forbidden when k=2, the failure processing is performed according to formula (6) to obtain the final input matrix

[0118]

[0119] Step 3.3: The subgraph features in are taken as nodes in the global interaction graph GNN:

[0120]

[0121] Among them, GNN(·) is a graph neural network, which is implemented through the self-attention mechanism;

[0122] is the adjacency matrix, which represents the spatial distance between nodes;

[0123] To extract the global interaction features, Divided into historical input features and future real features

[0124] Step 4: Decode multiple possible trajectories through the decoder, and select the trajectory that best matches the driving intention as the final predicted trajectory, such as Figure 6 As shown;

[0125] Furthermore, step 4 is implemented by:

[0126] Step 4.1: From the probability distribution P that conforms to the Gaussian distribution z (P z The latent space variable z is sampled from the posterior probability distribution of the future trajectory of the vehicle obtained through training. i , and through a linear layer After matching the dimensions with Splicing As shown in formula (7), After decoding by the decoder, a trajectory s is obtained i ;

[0127]

[0128] Step 4.2: Repeat step 4.1 N times for each target vehicle to obtain a set of possible future trajectories. is the decoder, T pred represents the time step of the predicted trajectory;

[0129] Step 4.3: Determine the lane that the target vehicle will pass through based on its driving intention, and use the endpoint of the lane centerline as the reference point for screening future trajectories. Take 6 coordinate points at equal intervals for each decoded future trajectory, and calculate the Euclidean distance between each point and the reference point, and sum the results. The trajectory corresponding to the minimum sum is the final predicted trajectory.

[0130] Step 5: In the training phase, the data collected at the road intersections are used as samples for training;

[0131] Furthermore, step 5 is implemented by:

[0132] Step 5.1: and After concatenation, input an MLP layer;

[0133] Step 5.2: The result of step 5.1 is estimated by conditional variational autoencoder (CVAE) to be and the variance is The latent variable z i , is a Gaussian distribution. i and The reconstructed future trajectory is obtained as the input of the decoder LSTM network The training process uses the loss function (8) Minimize the error with the true future trajectory Y, where the first term is the mean square error loss, which is used to measure the Euclidean distance between the predicted value and the true value; the second term is the KL divergence, which is used to measure the closeness of the latent space variable z to the Gaussian distribution.

[0134]

[0135] Among them, q φ for For approximate fitting of standard Gaussian distribution cognitive network.

[0136] Step 6: Perform trajectory prediction on the roadside device and transmit the result to each vehicle;

[0137] Furthermore, step 6 is implemented by:

[0138] The road-side device predicts the trajectory of each target vehicle at the intersection, and all predicted trajectories are sent to each target vehicle by the communication device, and the predicted trajectory visualization results can be further sent to the vehicle display. The driver can intuitively understand the trajectory of surrounding traffic participants in the next 3 seconds from the display, thereby increasing the driver's trust in the trajectory prediction system and improving driving safety.

[0139] Usually, because the predicted trajectory is not necessarily completely accurate, the driver can intervene in the vehicle when he finds a low-level or obvious error, which can also be considered to increase the certainty of the entire solution.

[0140] The steps in this application can be adjusted in order, combined, and deleted according to actual needs.

[0141] The units in the device of the present application can be combined, divided and deleted according to actual needs.

[0142] Although the present application is disclosed in detail with reference to the accompanying drawings, it should be understood that these descriptions are merely exemplary and are not intended to limit the application of the present application. The scope of protection of the present application is defined by the appended claims and may include various modifications, alterations and equivalents made to the invention without departing from the scope and spirit of the present application.

Claims

1. A method for predicting intersection trajectories based on strategic intent, comprising the following steps: Step 1: Set up roadside equipment at the intersection, collect historical trajectory information of traffic participants including vehicles and road environment, obtain high-precision maps of the area, and obtain original data; Step 2: vectorize the scene, encode each constructed sub-graph, generate a global graph, and obtain global interaction features; Step 2 is achieved by: Step 2.1, divide the intersection into five areas, which are numbered k, k = 1, 2, 3, 4, 5; Step 2.2: Vectorize the information in each area, abstract each scene information including vehicle trajectory, road, and lane line into a polyline, and convert the vector V i The head and tail coordinates, semantic information, attribute number and region number of are represented as a two-dimensional matrix: Among them, V i The first column represents the starting point coordinates; The second column represents the end point coordinates; The third column represents the attribute and sampling frequency, i.e., the semantic label; The fourth column represents the same attribute number; The fifth column represents the region number; Step 2.3: V with the same attribute and the same i i Connect to form a line subgraph Step 2.4: Encode the line subgraph features. The encoding method is as follows: in, It represents the use of one-dimensional convolution to encode the input features; and Represents maximum pooling and mean pooling respectively; is a linear mapping; Step 3: construct an input matrix according to the divided areas, and invalidate the restricted areas according to traffic rules; Step 3.1: Use the line subgraph features encoded in step 2.4 to construct an input matrix according to the p subgraphs contained in the region k where it is located: Step 3.2: According to the no-travel restrictions obtained by the roadside equipment, the input matrix of the corresponding area is invalidated. The processing method is as follows: in, Represents the feature input of the entire intersection, which is composed of the above five input matrices, where 0 is the non-road area; Θ is the prohibited area identifier for each area, which is 0 when prohibited and 1 otherwise; When k=2 areas are forbidden, the failure processing is performed according to formula (5) to obtain the final input matrix Step 3.3: The subgraph features in are taken as nodes in the global interaction graph GNN: Among them, GNN(·) is a graph neural network, which is implemented through the self-attention mechanism; is the adjacency matrix, which represents the spatial distance between nodes; To extract the global interaction features, Divided into historical input features and future real features Step 4: Obtain multiple possible trajectories through decoding by the decoder, and select the trajectory that best meets the driving intention as the final predicted trajectory; Step 5: In the training phase, the data collected at the road intersections are used as samples for training; Step 6: Perform trajectory prediction on the road-side device and transmit the results to each vehicle.

2. The intersection trajectory prediction method based on strategic intention according to claim 1 is characterized in that: Step 1 is achieved by: Step 1.1, perform panoramic acquisition of the intersection with a sampling frequency of 10 Hz; Step 1.2, obtain the high-precision map of the intersection, and coordinate the scene information collected in step 1.1 on the high-precision map; Step 1.3, the road-side device receives in advance the strategic intention sent by the vehicle through the sending module, thereby obtaining its traveling direction at the intersection, that is, the driving intention of the intersection.

3. The intersection trajectory prediction method based on strategic intention according to claim 2 is characterized in that: Step 4 is achieved by: Step 4.1: From the probability distribution P that conforms to the Gaussian distribution z Sample the latent space variable z in i , and through a linear layer After matching the dimensions with Splicing As shown in formula (7), After decoding by the decoder, a trajectory s is obtained i ; Step 4.2: Repeat step 4.1 N times for each target vehicle to obtain a set of possible future trajectories. is the decoder, T pred represents the time step of the predicted trajectory; Step 4.3: Determine the lane that the target vehicle will pass through based on its driving intention, and use the endpoint of the lane centerline as the reference point for screening future trajectories; take 6 coordinate points at equal intervals for each decoded future trajectory, and calculate the Euclidean distance between each point and the reference point, sum the results, and the trajectory corresponding to the minimum sum is the final predicted trajectory.

4. The intersection trajectory prediction method based on strategic intention according to claim 3 is characterized in that: Step 5 is achieved by: Step 5.1: and After concatenation, input an MLP layer; Step 5.2: The result of step 5.1 is estimated by conditional variational autoencoder (CVAE) to be and the variance is The latent variable z i , is a Gaussian distribution; z i and The reconstructed future trajectory is obtained as the input of the decoder LSTM network 5. The intersection trajectory prediction method based on strategic intention according to claim 4 is characterized in that: Step 6 is achieved by: The trajectory of each target vehicle at the intersection is predicted on the road-side device, and all predicted trajectories are sent to each target vehicle by the communication device. The predicted trajectory visualization results can be further sent to the vehicle display.

Images