A UAV-assisted blind spot pedestrian detection system for connected vehicles based on device-edge-cloud collaboration

Abstract

The present invention discloses a UAV-assisted vehicle network blind spot pedestrian detection system based on end-edge-cloud collaboration. The system includes at least one UAV terminal, a mobile edge node, a static edge node and a cloud node, and each node has computing and transmission capabilities. The method includes: 1. The UAV is deployed in the air to monitor the vehicle blind spot in real time, collect real-time video stream information, and promptly inform the static edge node on the road side of the blind spot pedestrian detection service; 2. The static edge node determines which node the task will be unloaded to according to a certain scheduling algorithm based on the task requirement information, the available computing resources and communication bandwidth of the heterogeneous nodes; 3. The static edge node notifies the UAV of the scheduling result, and the UAV sends the monitored video stream to the corresponding node through V2X communication; 4. The corresponding node performs the pedestrian detection task according to the deployed model, and sends a warning message to the vehicle when a potential collision risk is detected. The present invention provides a solution for assisting the vehicle network through a UAV, which realizes the prevention of dangerous collisions in the blind spots of vehicles in the traffic system.

Description

Technical Field

[0001] The present invention relates to drones, Internet of Vehicles, edge computing, task offloading, V2X communication and intelligent transportation systems, and in particular to a drone-assisted Internet of Vehicles blind spot pedestrian detection system based on end-edge-cloud collaboration and a processing method thereof. Background Art

[0002] In recent years, the development of wireless communication and intelligent network technology has promoted the development of vehicular ad hoc networks (VANETs), which are the basis of emerging intelligent transportation systems (ITS). On the other hand, with the popularization of drone-based applications such as environmental monitoring, intelligent surveillance, wireless communication, and aerial photography, it is a vision to form a new paradigm of VANET and drone cooperation to achieve innovative and powerful information technology services.

[0003] There has been a lot of research on the integration of UAVs with VANETs, ​​including data forwarding, traffic monitoring, computation offloading, and trajectory optimization. In the data forwarding scenario, with UAVs as relay nodes, data caching and trajectory planning problems are studied to maximize network throughput. In the traffic monitoring scenario, UAVs make full use of their mobility to monitor the traffic network status in real time and report abnormal conditions to the control center. In the computation offloading scenario, the research focuses on considering the communication, storage, and computational resource allocation costs between UAVs and VANETs, ​​reducing computational processing delays, and saving UAV energy consumption. In the trajectory optimization scenario, by planning the trajectory of UAVs, the maximum system throughput and reduced service delay when integrating UAVs with VANETs are studied.

[0004] In the current traffic system, in many cases, the blind spot in front of the vehicle is the main cause of traffic accidents, and the blind spot of the vehicle on the road can be fixed or randomly generated. For example, at a certain time and location, there is a large bus or obstacle in front of the vehicle, and the blind spot will be randomly generated in front of the vehicle. In this case, the vehicle is very likely to cause a traffic accident. At present, ITS mainly uses roadside surveillance cameras to broadcast real-time road conditions to eliminate this hidden danger, but the blind spot is generated randomly. This method cannot help vehicles detect and warn blind spots in real time in such a scenario, and the prevention effect of traffic accidents is not good. In addition, the traditional cloud computing architecture cannot provide real-time and efficient services for road vehicles. How to solve the randomness of the blind spot of the vehicle and solve the problems of untimely task response of the cloud computing architecture are still huge challenges. Summary of the invention

[0005] To solve the above problems, the present invention mainly provides a UAV-assisted vehicle-to-vehicle blind spot pedestrian detection system based on end-edge-cloud collaboration. Specifically, a UAV with computing power and communication interface is considered to be hovering in the air to monitor the blind spots that may appear on the road for vehicles, and pedestrians can be detected based on a certain target detection model. At the same time, the UAV can communicate with vehicles, roadside infrastructure and cloud servers through V2V (vehicle-to-vehicle), V2I (vehicle-to-infrastructure) and V2C (vehicle-to-cloud) communications respectively. Therefore, the UAV can offload the target detection task to mobile vehicles, static edge nodes deployed on the roadside and cloud servers by transmitting the monitored video to the corresponding nodes. Due to the high flexibility of the UAV, the system solves the problem of randomness in the blind spots of the vehicle's field of view on the one hand, and overcomes the problem of slow cloud computing response on the other hand by adopting the end-edge-cloud collaboration method.

[0006] In order to solve the above technical problems, the main steps of the UAV-assisted vehicle network blind spot pedestrian detection system based on end-edge-cloud collaboration provided by the present invention are as follows:

[0007] Step 1: Offline model training for real-time monitoring of pedestrians in blind spots. Collect pedestrian data sets taken by drones for model training, and deploy the trained models on terminals, mobile edges, static edges, and cloud nodes with heterogeneous computing and communication capabilities. The present invention considers using two classic target detection models, SSD and YOLO. Compared with SSD, YOLO is a larger network that requires higher computational overhead and has better detection effects; SSD is a single-stage network that simultaneously implements object detection and classification. The lightweight and fast inference characteristics of SSD make it more suitable for mobile devices.

[0008] The training dataset of the system comes from a video of a university campus square taken by a drone. 2000 frames are randomly selected from the video, 90% of which are used for training and the remaining 10% are used for model verification. The batch size is set to 32, the IOU threshold is set to 0.98, and the training iteration is 5000. The two models are tested in a traffic scene. The drone video frames are collected, and the scene simulates the potential blind spots of the vehicle. Finally, the experimental results show that the detection accuracy of YOLO and SSD can be maintained at 95% and 90% respectively.

[0009] Step 2: Deploy the drone in the air to monitor the blind spots of vehicles on the traffic road, start the blind spot pedestrian detection service and notify the static edge nodes on the roadside. After receiving the notification, the static edge nodes start to update the task offloading strategy online. Specifically, the drone uses FFmpeg to implement the video decoding operation, and converts the decoded results into images in the YUVImage format to achieve image transmission and storage. The final YUV video stream is sent to each node in the system through V2X communication.

[0010] Step 3: A method is provided for determining to offload the task to a certain node for execution by combining the computing resources available at the heterogeneous nodes and the communication bandwidth of the network. The method comprises:

[0011] Step 3.1: Under the constraints of node computing, storage and communication resources, define the objective function with minimizing the average task delay as the optimization goal:

[0012] The node parameters include: define S = {s1, s2, ..., s |S|}, M={m1,m2,…,m |M|}, U={u1,u2,…,u |U| {c} and {d} are respectively a collection of static edge nodes, mobile edge nodes, terminal nodes and cloud servers;

[0013] Let N = {s1, s2, …, s |S| ,m1,m2,…,m |M| ,u1,u2,…,u |U| ,c} represents the uninstall node;

[0014] UAV perception task set express;

[0015] Each task can be represented as a 5-tuple They represent the input data size, the number of CPU cycles required, the accuracy requirement, the generation time and the deadline respectively;

[0016] Each unloading node n∈N is associated with 5 tuples <C n ,B n ,ρ n ,S n ,R n >, respectively representing the radius of communication coverage, total wireless bandwidth, number of channels, maximum storage function, and computing power (e.g., number of CPU cycles per unit time);

[0017] Remember u,n,t =1 means u∈N can exchange messages with n∈N in the single-hop communication at time t;

[0018] Let H = {h1,h2,…,h |H|} represents the weight of the neural network model;

[0019] Assume that the neural network model h k The task w ji Unload to node n for reasoning and define a binary variable

[0020] Communication delay between nodes:

[0021]

[0022] The total transmission delay includes the UAV sending video data to the unloading node and the unloading node sending the calculation results to the vehicle. The total communication delay of task unloading is defined as follows:

[0023]

[0024] The computational latency of the task:

[0025]

[0026] Task waiting delay:

[0027]

[0028] Finally, w ji The latency of the task offloaded to node n is given by It means that it is the calculation of total transmission delay, calculation delay and waiting delay:

[0029]

[0030] The goal of the system is to minimize the average task delay by determining the offloading strategy. The objective function is:

[0031]

[0032] The constraints of the objective function include:

[0033] Constraints C1 and C2 indicate that each task must be offloaded and computed to one node:

[0034] C1:x j,i,n,h ∈{0,1}

[0035] C2:∑ n∈N x j,i,n,h =1

[0036] Constraint C3 represents the storage constraint:

[0037]

[0038] Constraint C4 indicates that the number of tasks transmitted simultaneously cannot exceed the number of channels of the offloading node:

[0039]

[0040] Constraint C5 indicates that the selected neural network model meets the accuracy requirements:

[0041]

[0042] Constraint C6 means that the task must be completed before the deadline:

[0043]

[0044] Step 3.2: According to the optimization goal of the objective function, a task offloading algorithm is designed:

[0045] The task latency is estimated based on the currently applied offloading strategy. If it exceeds a predefined threshold, a greedy method is used to search for a new offloading strategy. It consists of two strategies, one is the latency-driven strategy, which aims to minimize the overall task latency of the system; the other is the resource-driven strategy, which aims to maximize resource utilization.

[0046] Delay-driven strategy: Given a certain offloading strategy, the task delay is t cur When the data is θ ji and the calculation requirement is c ji When a new task arrives, the unloading delay t is estimated according to the delay model derived above. aver When t cur and t aver The difference exceeds the predefined threshold t diff When , the algorithm starts to look for new offloading strategies. Specifically, it traverses each computing node and each neural network model and calculates the total task latency Then, select Minimal strategy.

[0047] Resource-driven strategy: Consider maximizing resource utilization, taking computing resource availability as an example. Specifically, given an offloading strategy, the computing latency is When the R n When it increases, the algorithm starts to look for a new offloading strategy. The specific update strategy is similar to the task-driven delay strategy.

[0048] Step 4: The drone communicates with other nodes through V2X. Specifically, the drone, static edge nodes, and mobile edge nodes are all located in a connected vehicle network and communicate through methods such as WiFi and DSRC. The drone communicates with the cloud node through 4G, 5G or cellular networks. The drone can receive the task offloading node selection results from the static edge nodes and send the collected blind spot video stream to the corresponding node for task calculation.

[0049] Step 5: Based on the pedestrian detection model obtained in step 1, the node of the computing task will deploy the offline trained pedestrian detection model in advance. The node will select the corresponding deep learning model for inference calculation according to the received message. The selection result can be obtained by the two task offloading strategies mentioned in step 3.2. After the node obtains the calculation result, considering the limited communication bandwidth and the validity of the result, it will not send all the calculation results to the vehicles on the road, but will only send a warning signal to the vehicle when a collision risk is detected. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] The accompanying drawings of the present invention are as follows:

[0051] Figure 1 is a system schematic diagram of the present invention;

[0052] Figure 2 is a flow chart of the present invention; DETAILED DESCRIPTION

[0053] Figure 1 It is a system schematic diagram of the present invention. The diagram mainly shows that the terminal-edge-cloud collaborative architecture consists of four basic elements: a drone terminal, a vehicle as a mobile edge node, a roadside unit as a static edge node, and a cloud server. Each element has certain computing and communication capabilities, so that the drone terminal can communicate with the vehicle, the roadside unit, and the cloud through V2X communication. Under this architecture, the general application scenario is described as follows: The drone is deployed in the air and can monitor the blind spots of the driving vehicle, such as the pedestrians surrounded by circles in the figure. The monitored video stream can be processed according to a specific object detection model (such as YOLO). The model can be deployed in drones, vehicles, roadside units, and clouds for pedestrian detection. The roadside unit obtains the unloading node of the current task according to a certain unloading algorithm and notifies the drone. The drone end transmits the video stream to the corresponding node through V2X communication for task calculation. Finally, if a potential pedestrian collision is detected, the warning information is transmitted to the driving vehicle through the task calculation node.

[0054] Figure 2 It is a specific implementation flow chart of the present invention. The following will describe the system execution steps in detail:

[0055] In step 101, the pedestrian detection model is trained offline, and two classic models, SSD and YOLO, are used for training. The lightweight feature of SSD is more suitable for mobile devices, and YOLO has higher detection accuracy. A large number of pedestrian data sets taken by drones in the air on traffic roads are collected, and the trained models are deployed on drone terminals, roadside static edge nodes, vehicle-mounted mobile edge nodes, and remote cloud nodes.

[0056] In step 102, a drone is deployed in the air above a traffic road. Since most drones do not have computing capabilities, a Raspberry Pi can be mounted on the drone as a processor to monitor the blind spots of vehicles on the traffic road.

[0057] In step 103, the drone monitors the blind spot conditions in real time, captures information such as the location of pedestrians, and then uses FFmpeg to implement video decoding operations, converting the decoded results into YUVImage format images to achieve video stream transmission and storage, and sends a notification to the roadside unit when starting to monitor the blind spot.

[0058] In step 104, the UAV communicates with the roadside unit through V2I. There are many channel interference signals in the traffic road, and the packet loss phenomenon is relatively serious during the transmission process. The use of reliable transmission TCP protocol will greatly increase the transmission delay, which does not meet the requirements of the task response rate in the actual traffic road. Therefore, the present invention is mainly designed to use UDP broadcast to transmit data. While ignoring the loss of a few packets, the message or data is sent to the roadside unit. Here, the notification of the activation of the blind spot monitoring service is sent to the roadside unit.

[0059] In step 105, the roadside unit receives the notification information sent by the drone, and the roadside unit starts to update the task offloading strategy in real time. Since the computing power of the roadside unit is stronger than that of the vehicle-mounted terminal and the drone terminal, and it is closer to the actual traffic road than the cloud node, the present invention deploys the task offloading algorithm to the roadside unit in advance.

[0060] In step 106, the RSU integrates the computing, storage and bandwidth resources of all heterogeneous nodes in the system, determines to minimize the average task delay as the optimization goal, and defines the objective function, where the definition and constraints are as follows:

[0061] Related parameters include: Definition S = {s1, s2, ..., s |S|}, M={m1,m2,…,m |M|}, U={u1,u2,…,u |U| {c} and {d} are respectively a collection of static edge nodes, mobile edge nodes, terminal nodes and cloud servers;

[0062] Let N = {s1, s2, …, s|S| ,m1,m2,…,m |M| ,u1,u2,…,u |U| ,c} represents the uninstall node;

[0063] UAV perception task set express;

[0064] Each task can be represented as a 5-tuple They represent the input data size, the number of CPU cycles required, the accuracy requirement, the generation time and the deadline respectively;

[0065] Each unloading node n∈N is associated with 5 tuples <C n ,B n ,ρ n ,S n ,R n >, respectively representing the radius of communication coverage, total wireless bandwidth, number of channels, maximum storage function, and computing power (e.g., number of CPU cycles per unit time);

[0066] Remember u,n,t =1 means u∈N can exchange messages with n∈N in the single-hop communication at time t;

[0067] Let H = {h1,h2,…,h |H|} represents the weight of the neural network model;

[0068] Assume that the neural network model h k The task w ji Unload to node n for reasoning and define a binary variable

[0069]

[0070] Communication delay between nodes:

[0071]

[0072] The total transmission delay includes the UAV sending video data to the unloading node and the unloading node sending the calculation results to the vehicle. The total communication delay of task unloading is defined as follows:

[0073]

[0074] The computational latency of the task:

[0075]

[0076] Task waiting delay:

[0077]

[0078] Finally, w ji The latency of the task offloaded to node n is given by It means that it is the calculation of total transmission delay, calculation delay and waiting delay:

[0079]

[0080] The goal of the system is to minimize the average task delay by determining the offloading strategy. The objective function is:

[0081]

[0082] The constraints of the objective function include:

[0083] Constraints C1 and C2 indicate that each task must be offloaded and computed to one node:

[0084] C1:x j,i,n,h ∈{0,1}

[0085] C2:∑ n∈N x j,i,n,h =1

[0086] Constraint C3 represents the storage constraint:

[0087]

[0088] Constraint C4 indicates that the number of tasks transmitted simultaneously cannot exceed the number of channels of the offloading node:

[0089]

[0090] Constraint C5 indicates that the selected neural network model meets the accuracy requirements:

[0091]

[0092] Constraint C6 means that the task must be completed before the deadline:

[0093]

[0094] In step 107, the system estimates the task delay according to the currently applied offloading strategy. If it exceeds the predefined threshold, a greedy method is used to search for a new offloading strategy. It consists of two strategies, one is the delay-driven strategy, which aims to minimize the overall task delay of the system; the other is the resource-driven strategy, which aims to maximize resource utilization.

[0095] Given a certain offloading strategy, the system initially offloads tasks according to the offloading strategy, indicating that the task delay is t cur When the data is θji and the calculation requirement is c ji When a new task arrives, the unloading delay t is estimated according to the delay model derived above. aver The same resource-driven strategy considers maximizing resource utilization, taking computing resource availability as an example. Specifically, given an offloading strategy, the computing latency is When the R n When increasing, calculate the current unloading delay t aver ,The specific update strategy is similar to the task-driven delay strategy.

[0096] In step 108, when t cur and t aver The difference exceeds the predefined threshold t diff , the algorithm starts looking for a new unloading strategy.

[0097] In step 109, the RSU starts to look for the optimal task offloading strategy in the current environment. Specifically, it traverses each computing node and each neural network model to calculate the total task delay. Then, choose the greedy option The minimum strategy is used, and the unloading strategy is recorded and sent to the drone end.

[0098] In step 110, when t cur and t aver If the difference does not exceed the threshold, the algorithm continues to use the previous unloading strategy.

[0099] In step 111, the drone receives the unloading strategy notification from the roadside unit, and then sends the collected blind spot video stream to the corresponding node for task calculation. In this process, the drone communicates with other nodes through V2X. Specifically, the drone communicates with the static edge node and the mobile edge node in a vehicle network at the same time, such as WiFi, DSRC, etc. The drone communicates with the cloud node through 4G, 5G or cellular network.

[0100] In step 112, the node receives the video stream from the drone, and the node of the computing task will call the pre-deployed offline trained pedestrian detection model. The node will select the corresponding deep learning model for inference calculation based on the received message. After the node obtains the calculation result, considering the limited communication bandwidth and the validity of the result, it will not send all the calculation results to the vehicles on the road, but will only send a warning signal to the vehicle when a pedestrian collision risk is detected.

[0101] In step 113, the computing node detects the risk of pedestrian collision and sends a warning message to the vehicle. The driver in the vehicle takes corresponding actions based on the warning message and chooses whether to end the blind spot pedestrian detection service. If the service is not ended, the system continues to execute step 103.

[0102] In step 114, the computing node does not detect the presence of a pedestrian collision hazard, does not send a warning message, and the system will continue to perform the blind spot pedestrian detection service.

Claims

1. A drone-assisted vehicle-to-vehicle blind spot pedestrian detection system based on device-edge-cloud collaboration, characterized in that: The following steps are involved: Step 1: In the offline preparation phase, a data set containing pedestrians in dangerous sections of traffic roads is collected for training pedestrian detection models such as YOLO and SSD. The trained models can be deployed on terminals, mobile edges, static edges, and cloud nodes with heterogeneous computing and communication capabilities. Step 2: Deploy the drone in the air to monitor the blind spots of vehicles on the traffic road, start the blind spot pedestrian detection service and notify the static edge nodes on the roadside; Step 3: The roadside static edge node determines the task offloading strategy based on the available computing resources of all heterogeneous nodes in the current system, the communication bandwidth of the network, and the task requirements, and offloads the task to the corresponding node to perform the reasoning task. Specifically, task offloading consists of two strategies: one is the delay-driven strategy, which aims to minimize the overall task delay of the system, and the other is the resource-driven strategy, which aims to maximize resource utilization. Delay-driven strategy: Given a certain offloading strategy, the task delay is expressed as t cur , when the data is θ ji and the calculation requirement is c ji When a new task arrives, the task offloading delay t is estimated based on the node status and task requirements. aver , when t cur and t aver The difference exceeds the predefined threshold t diff When , the algorithm adopts a greedy method to search for new offloading strategies. Specifically, it traverses each computing node and each neural network model and calculates the total task delay Then, select Minimal strategy; Resource-driven strategy: Considering the maximization of computing resource utilization, given an offloading strategy, the computing latency is expressed as When the R n When it increases, the algorithm starts to look for a new offloading strategy. The specific update strategy is similar to the task-driven delay strategy; Step 4: The roadside static edge node transmits the task offloading result to the drone, and the drone sends the monitored video stream to the corresponding node through V2X communication; Step 5: The node that obtains the computing task detects and calculates the pedestrian according to the pedestrian detection model obtained in step 1. If a potential collision risk is detected in the blind spot, a warning message is sent to the vehicles on the road. The vehicles that receive the warning signal perform corresponding risk avoidance operations according to the actual driving situation.

2. The UAV-assisted vehicle-to-vehicle blind spot pedestrian detection system based on device-edge-cloud collaboration according to claim 1 is characterized by: In step 1, according to the actual situation of the traffic system, and to improve the accuracy of pedestrian detection in blind spots, this system considers using two classic target detection models, SSD and YOLO. Compared with SSD, YOLO is a larger network, requires higher computational overhead, and has better detection effect; SSD's lightweight and fast reasoning characteristics make it more suitable for mobile devices; The system's training data set comes from a video stream taken by a drone of a university campus square. 2,000 frames are randomly selected from the video, 90% of which are used for training and the remaining 10% are used for model verification. During the training process, the batch size is set to 32, the IOU threshold is set to 0.98, and the training iterations are 5,000. Finally, according to experimental tests, the detection accuracy of YOLO and SSD can be maintained at 95% and 90%, respectively.

3. According to the device-edge-cloud collaborative UAV-assisted vehicle network blind spot pedestrian detection system according to claim 1, the UAV in step 2 monitors the blind spot in real time, characterized in that: The method for acquiring video stream data from a drone is that the drone uses FFmpeg to implement video decoding operations and converts the decoding results into images in the YUVImage format to achieve image transmission and storage. The final YUV video stream is sent to each node in the system through V2X communication.

4. According to the device-edge-cloud collaborative UAV-assisted vehicle-to-vehicle blind spot pedestrian detection system of claim 1, the integration of computing and bandwidth resources available at heterogeneous nodes in step 3 is characterized by: The node parameters include: defining S = {s1, s2, ..., s |S| }, M={m1,m2,…,m |M| }, U={u1,u2,…,u |U| } and {c} are the sets of static edge nodes, mobile edge nodes, terminal nodes, and cloud servers, respectively; With N = {s1, s2, ..., s |S| ,m1,m2,…,m |M| ,u1,u2,…,u |U| ,c} represents the uninstall node; UAV perception task set express; Each task requirement is represented as a 5-tuple They represent the input data size, the number of CPU cycles required, the accuracy requirement, the generation time and the deadline respectively; Each unloading node n∈N uses 5 tuples <C n ,B n ,ρ n ,S n ,R n >Indicates its status, representing the radius of communication coverage, total wireless bandwidth, number of channels, maximum storage function, and computing power; Remember u,n,t =1 means u∈N can exchange messages with n∈N in the single-hop communication at time t; Let H = {h1,h2,…,h |H| } represents the weight of the neural network model; Assume that the neural network model h k The task w ji Unload to node n for reasoning and define a binary variable Communication delay between nodes: Total communication delay of task offloading: The computational latency of the task: Task waiting delay: Finally, w ji The task unloading delay to node n is given by It is the sum of total transmission delay, computation delay and waiting delay: The goal of the system is to minimize the average task delay by determining the offloading strategy. The objective function is as follows:

5. According to claim 4, a drone-assisted vehicle-to-vehicle blind spot pedestrian detection system based on device-edge-cloud collaboration, the objective function defined with average task delay as the optimization target is characterized by: The constraints of the objective function include: C1:x j,i,n,h ∈{0,1} C2:∑ n∈N x j,i,n,h =1 C3: C4: C5: C6: Among them, constraints C1 and C2 mean that each task must be offloaded and calculated to one node, constraint C3 represents the storage constraint, constraint C4 means that the number of tasks transmitted simultaneously cannot exceed the number of channels of the offloading node, constraint C5 means that the selected neural network model meets the accuracy requirements, and constraint C6 means that the task must be completed before the deadline.

6. According to the device-edge-cloud collaborative UAV-assisted vehicle network blind spot pedestrian detection system of claim 1, the communication between the UAV and other nodes in the system in step 4 is characterized in that: Drones communicate with other nodes through V2X. Specifically, drones, static edge nodes, and mobile edge nodes are all located in a connected vehicle network and communicate through methods such as WiFi and DSRC. Drones communicate with cloud nodes through 4G, 5G or cellular networks. Drones can receive task offloading node selection results from static edge nodes and send the collected blind spot video streams to the corresponding nodes for task calculation.

7. According to the device-edge-cloud collaborative UAV-assisted vehicle network blind spot pedestrian detection system of claim 1, the node in step 5 detects pedestrians in the blind spot, characterized in that: The nodes of the computing task will deploy the offline trained pedestrian detection model in advance. The nodes will select the corresponding deep learning model for inference calculation based on the received message strategy. After obtaining the calculation results, considering the limited communication bandwidth and the validity of the results, not all calculation results will be sent to the vehicles on the road. Only a warning signal will be sent to the vehicle when a collision risk is detected.

Images