Road-Cloud Collaborative Pedestrian Prediction-Guided Autonomous Driving Path Planning System

Abstract

The present invention discloses an autonomous driving path planning system with road-cloud collaboration and pedestrian prediction guidance, which belongs to the field of autonomous driving technology; by modeling pedestrian behavior characteristics, pedestrian movement trajectories in a certain future time axis are predicted, and this information is used to plan the future movement trajectory of autonomous driving vehicles. The pedestrian trajectory feature model is transplanted into the edge cloud device, combined with millimeter wave radar and camera equipment to construct a two-dimensional particle space, and the trajectory of the particle target in the future is predicted based on the real-time pedestrian movement trajectory, so as to plan in advance a shorter and more comfortable autonomous driving vehicle driving path in the future, while empowering traffic safety and providing better autonomous driving services or non-autonomous driving assistance services.

Description

Technical Field

[0001] The present invention relates to the field of autonomous driving technology, and in particular to an autonomous driving path planning system with road-cloud collaboration and pedestrian prediction guidance. Background Art

[0002] As autonomous driving develops further, the amount of data generated by sensors will increase exponentially. An autonomous vehicle can generate up to 25TB of raw data per day. In addition to the difficulties caused by storage, it is also necessary to consider that these raw data need to be processed, integrated, and reprocessed in real time in the car, which places great demands on the processor's capabilities and energy consumption. Since these computing tasks need to be terminated in the car to ensure the real-time processing and response, a powerful and reliable edge computing platform is required to execute them. The traditional TSP (Telematics Service Provider) platform can no longer meet the continued development of autonomous driving. Summary of the invention

[0003] The purpose of the present invention is to address the above-mentioned problems and to provide an edge cloud device based on pedestrian trajectory prediction, in which a road-cloud-coordinated pedestrian prediction-guided autonomous driving path planning system is pre-installed. The system predicts the trajectory of the particle target in the future based on the real-time acquired pedestrian motion trajectory, thereby planning in advance a shorter and more comfortable autonomous driving vehicle driving path in the future.

[0004] In order to achieve the above object, the technical solution adopted by the present invention is:

[0005] An autonomous driving path planning system with road-cloud collaboration and pedestrian prediction guidance includes the following contents:

[0006] Pedestrian trajectory acquisition module: used to obtain the image behavior information and actual position information of pedestrians, and fuse them to obtain the actual movement trajectory of pedestrians;

[0007] Behavior prediction module: used to import the actual motion trajectory of multiple frames of pedestrians into the trained pedestrian motion trajectory prediction model to generate the pedestrian predicted motion trajectory;

[0008] Reliable determination module: used to obtain multiple segments of pedestrian predicted motion trajectories and pedestrian actual motion trajectories in the previous period of the pedestrian predicted motion trajectory, and analyze and calculate their similarities to obtain the final value of the overlap; if the final value of the overlap is greater than or equal to the preset threshold, the prediction is determined to be reasonable, otherwise it is determined to be a prediction failure;

[0009] Vehicle trajectory planning module: used to import the predicted reasonable pedestrian motion trajectory into the vehicle trajectory planning model, predict whether there will be a collision with the pedestrian based on the current motion speed and posture, and generate the vehicle motion trajectory;

[0010] Result output module: used to respond to the request of the on-board lower computer and send the vehicle motion trajectory or pedestrian predicted motion trajectory prediction failure information to the on-board lower computer through the MEC communication module.

[0011] Among them, the specific operations of the pedestrian trajectory acquisition module are: through the calibration algorithm, the millimeter-wave radar and camera group are calibrated to the same coordinate system and associated with the world coordinate system; the millimeter-wave radar data is clustered by the clustering algorithm to cluster the collected point information of the person into particles to obtain the relative actual position information of the pedestrian; the camera module is subjected to target detection at a certain frame rate, and the target detection and tracking are realized through the Yolo algorithm to obtain the image behavior information of the pedestrian; the obtained image behavior information and the relative actual position information are integrated to establish a two-dimensional particle coordinate map to obtain the actual motion trajectory of the pedestrian with position information.

[0012] The specific operation of the behavior prediction module is as follows: first, the position information obtained by fusing multiple frames of pedestrians’ actual motion trajectories is processed; then the information from multiple frames is combined and imported into the trained pedestrian motion trajectory prediction model; and then the pedestrian motion trajectory prediction model generates a predicted motion trajectory route and reflects it in a two-dimensional particle coordinate diagram.

[0013] The specific operation of the reliable judgment module is as follows: obtaining the actual trajectory A of the nth segment, the predicted trajectory B of the Nth segment, and the predicted trajectory C of the N+1th segment of the pedestrian's predicted motion trajectory, analyzing the similarity between the predicted trajectory C of the N+1th segment and the actual trajectory A of the nth segment and the predicted trajectory B of the Nth segment to obtain two overlaps, and summing the two overlaps according to the weight ratio trained by the neural network model to obtain a final overlap value. If the final overlap value is greater than or equal to a preset threshold, the prediction is judged to be reasonable, otherwise it is judged to be a prediction failure; the preset threshold is 90%.

[0014] In a preferred embodiment, the road-cloud coordinated pedestrian prediction-guided autonomous driving path planning system also includes a screening module, which is used to divide the road into a number of lanes, and use the first pedestrian motion trajectory that intersects with the lane in the upward and downward directions of each lane as a screening condition, and screen out the predicted pedestrian motion trajectory that needs to be judged for reliability from all the pedestrian motion trajectories within a certain period of time within the detection range, and output it to the reliable judgment module for reliability judgment, wherein the pedestrian motion trajectory includes the actual pedestrian motion trajectory and the predicted pedestrian motion trajectory.

[0015] Due to the adoption of the above technical solution, the present invention has the following beneficial effects:

[0016] 1. The present invention predicts the movement trajectory of pedestrians in a certain time axis in the future by modeling pedestrian behavior characteristics, and uses this information to plan the future movement trajectory of autonomous vehicles. The pedestrian trajectory feature model is transplanted into the edge cloud device, combined with millimeter wave radar and camera equipment to construct a two-dimensional particle space, and the trajectory of the particle target in the future is predicted based on the real-time pedestrian movement trajectory, so as to plan in advance a shorter and more comfortable autonomous driving vehicle driving path in the future, while empowering traffic safety and providing better autonomous driving services or non-autonomous driving assistance services.

[0017] 2. The present invention designs a predicted trajectory reliability determination module, which compares the trajectory overlap between the previous and next frames. When the contrast exceeds a certain threshold, the predicted trajectory is considered reasonable, thus ensuring the reliability of the decision. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 It is a processing flow chart of the present invention.

[0019] Figure 2 It is an application layer block diagram of the present invention.

[0020] Figure 3 It is the two-dimensional particle coordinates and the particle target trajectory diagram after clustering of the present invention.

[0021] Figure 4 It is a pedestrian motion trajectory prediction model diagram of the present invention.

[0022] Figure 5 It is a trajectory similarity determination model diagram of the present invention. DETAILED DESCRIPTION

[0023] The specific implementation of the invention is further described below with reference to the accompanying drawings.

[0024] Example 1

[0025] like Figure 1 As shown, a road-cloud-coordinated pedestrian prediction-guided autonomous driving path planning system and its specific processing flow include the following contents:

[0026] Pedestrian trajectory acquisition module: used to obtain the image behavior information and actual position information of pedestrians, and fuse them to obtain the actual movement trajectory of pedestrians. The specific operations are as follows: through the calibration algorithm, the millimeter wave radar and camera group are calibrated to the same coordinate system and associated with the world coordinate system; the millimeter wave radar data is clustered by the clustering algorithm to cluster the collected point information of people into particles to obtain the relative actual position information of pedestrians; the camera module is subjected to target detection at a certain frame rate, and the target detection and tracking are realized through the yolo algorithm to obtain the image behavior information of pedestrians; the obtained image behavior information and the relative actual position information are fused to establish a two-dimensional particle coordinate map to obtain the actual movement trajectory of pedestrians with position information.

[0027] Behavior prediction module: used to import the actual motion trajectory of multiple frames of pedestrians into the trained pedestrian motion trajectory prediction model to generate the predicted motion trajectory of pedestrians. The specific operation is: first, process the position information obtained by fusing the actual motion trajectory of multiple frames of pedestrians; then combine the information under multiple frames and import it into the trained pedestrian motion trajectory prediction model; then generate the predicted motion trajectory route from the pedestrian motion trajectory prediction model and reflect it in a two-dimensional particle coordinate diagram.

[0028] Reliable judgment module: used to obtain multiple segments of the pedestrian's predicted motion trajectory and the pedestrian's actual motion trajectory in the previous period of time, and analyze and calculate their similarities to obtain the final value of the overlap; if the final value of the overlap is greater than or equal to the preset threshold, the prediction is judged to be reasonable, otherwise it is judged to be a prediction failure. The specific operation is: obtain the actual trajectory A of the nth segment of the pedestrian's predicted motion trajectory, the predicted trajectory B of the Nth segment, and the predicted trajectory C of the N+1th segment, analyze the similarity of the predicted trajectory C of the N+1th segment with the actual trajectory A of the nth segment and the predicted trajectory B of the Nth segment to obtain two overlaps, and sum the two overlaps according to the weight ratio trained by the neural network model to obtain the final value of the overlap, if the final value of the overlap is greater than or equal to the preset threshold, the prediction is judged to be reasonable, otherwise it is judged to be a prediction failure.

[0029] Vehicle trajectory planning module: used to import the predicted reasonable pedestrian motion trajectory into the vehicle trajectory planning model, predict whether there will be a collision with the pedestrian based on the current motion speed and posture, and generate the vehicle motion trajectory.

[0030] Result output module: used to respond to the request of the on-board lower computer and send the vehicle motion trajectory or pedestrian predicted motion trajectory prediction failure information to the on-board lower computer through the MEC communication module.

[0031] The following will be explained in detail in conjunction with relevant equipment.

[0032] The present invention aims to solve the future trajectory planning of autonomous vehicles under general traffic intersection conditions. With edge cloud devices (MEC) as the intermediate machine, the behavior prediction algorithm (AIP), cloud service allocation algorithm (AIC), multi-sensor fusion algorithm, prediction result reliability judgment model and other algorithms and models are deployed to obtain specific trajectory results and extend the prediction range of the motion trajectory of autonomous vehicles. By analyzing the data provided by the actual camera and millimeter wave radar, a two-dimensional particle map is established to predict the motion trajectory of pedestrians in the future, and based on the prediction results, the trajectory of the driverless car is optimized. All information interactions and the use of MEC's ​​own computing power are scheduled through the cloud service sub-node intelligent allocation algorithm based on the event trigger mechanism (when the information priority is high, more computing power will be dispatched for calculation, see below), and finally achieve safer traffic conditions, maximized use of the equipment's own computing power, optimal vehicle driving path, and optimal user comfort.

[0033] like Figure 2 As shown, the core service layer of the present invention can be divided into a user interface layer; a MEC service layer; a hardware device layer; an algorithm service layer; and a data layer.

[0034] The user interface layer opens some API interfaces for customers, and provides real-time data information and entertainment services through MEC. The main service objects are autonomous vehicles and passengers. The technology used is real-time reception and transmission. In order to meet high concurrency and real-time requirements, we transplanted and called the CyberRT module under the APOLLO autonomous driving framework. As the autonomous driving framework of Baidu APOLLO, the component functions in CyberRT can meet our needs for real-time data transmission and reception. Compared with the message transmission and reception provided by the traditional ROS2 framework. CyberRT has better real-time performance and has been expanded in three aspects: communication function optimization, decentralized network topology, and data compatibility expansion (CyberRT is a product of Baidu APOLLO, which has these features).

[0035] The MEC business layer involves four algorithm modules: MEC resource scheduling communication module (MEC communication module), MEC short-distance path planning module (MEC trajectory planning model), MEC pedestrian behavior trajectory prediction module (MEC pedestrian trajectory prediction module), and MEC perception data fusion processing module (MEC data processing module). These modules run through the entire life cycle of the MEC device. Through the CyberRT framework, the perception data fusion module receives and processes real-time information collected from hardware devices, and publishes the processed trajectory information to the pedestrian behavior trajectory prediction module. The module generates pedestrian movement information on a future timeline through a pre-trained pedestrian behavior feature model. This information is transmitted to the short-distance path planning module, which mainly serves autonomous driving vehicles. Through the pedestrian trajectory on a certain future timeline provided by the prediction module, the short-distance planning module plans the speed and acceleration in advance. Generate a path trajectory with the smallest Jerk value (an indicator for evaluating the comfort of automobile movement) and the shortest path that will not cause a collision. The most important function of the final MEC resource scheduling communication module is the transmission of trajectory information. In addition to basic message transmission, this module uses a task scheduling algorithm to assign task priorities to prevent message delays and idle computing power, and to give full play to the capabilities of MEC devices.

[0036] The hardware device layer is mainly composed of camera modules and millimeter-wave radars. The camera provides portrait information and realizes target tracking through the YOLO algorithm, while the millimeter-wave radar provides location information and clusters into points through the clustering algorithm. By fusing the data of the two devices, a two-dimensional particle map is established, the motion trajectory of the particle target is drawn, and information such as the target type and identity is given.

[0037] The data layer provides services for temporary storage and call, temporarily storing prediction results and fused data sets, and can also be used as a storage area for general monitoring equipment. The purpose of this layer is to store information and facilitate call. The performance indicators of the evaluation depend on the selection of hardware.

[0038] The present invention is deployed at general traffic intersections, and the traffic participating units involved include: various types of autonomous driving vehicles, general non-autonomous driving vehicles, and pedestrians. MEC equipment is installed on the ground base station, the camera module and the millimeter wave radar are deployed within a range of 4 to 5 meters from the ground, with a depression angle of 30° to obtain the best data measurement range. The corresponding electrical equipment is deployed near the base station, and the equipment is powered by AC power. The relevant processing flow is as follows:

[0039] Multi-sensor fusion algorithm: The purpose of fusion is to process the point cloud data obtained by the millimeter-wave radar and the two-dimensional image obtained by the camera group in the actual scene to fuse the data, so as to obtain richer target information. This algorithm mainly strengthens the acquisition of relative distance and target information itself. 1. The algorithm uses the calibration algorithm to calibrate the millimeter-wave radar and the camera group to the same coordinate system and associate them with the world coordinate system. 2. For the millimeter-wave radar data, an improved DPC (clustering by fast search and find of density peak) clustering algorithm is performed, and the collected millimeter-wave radar data of people and objects are clustered into particles to obtain accurate relative position information. 3. The camera module performs target detection at a certain frame rate (30fps), and the target detection and tracking is achieved through the yolo algorithm. The algorithm realizes target detection and tracking. 4. The obtained data information is fused to establish a two-dimensional particle coordinate map. Finally, the target trajectory with position information is obtained. See Figure 3 .

[0040] Behavior Prediction Algorithm (AIP): As the core module of MEC equipment, the prediction algorithm realizes the prediction of pedestrian motion trajectory over a certain period of time. 1. The algorithm first processes the position information obtained by the fusion algorithm under multiple frames (10 frames). 2. Combine the information under multiple frames and import it into the trained pedestrian motion trajectory prediction model (LSTM model or improved LSTM model, both of which are existing technologies). 3. Generate the predicted trajectory route and reflect it in a two-dimensional particle coordinate diagram. See Figure 4 .

[0041] Reliability Judgment Module (RJM): The reliability of the prediction results is extremely important for traffic safety, so a separate module is needed as an independent judgment mechanism to make feasibility judgments. RJM is a judgment mechanism for whether the prediction results are reliable. This module relies on the prediction trajectory results within one second ( Figure 3), it analyzes whether the final value of the overlap between the result trajectory B of the Nth segment and the result trajectory A of the nth segment, and the trajectory C of the N+1 segment is greater than the threshold; specifically, the result of the N+1th frame is compared with the Nth frame for similarity, and the result of the N+1th frame is compared with the nth frame for similarity, and the values ​​describing the overlap are obtained respectively, and the two overlaps are summed according to the weight ratio trained by the neural network model. If the summed result (the final value of the overlap) is higher than a certain threshold, it passes to prevent the predicted result from not converging. If the final value of the overlap meets expectations (such as the preset threshold of the final value of the overlap is greater than or equal to 90%), the predicted trajectory passes, otherwise it fails. The typical values ​​of the preset threshold are 85%, 90% or 95%, etc. The module compares the overlap of the predicted trajectories in the previous and next frames. When the contrast exceeds a certain threshold, the predicted trajectory can be considered reasonable to ensure the reliability of the decision. See Figure 5 .

[0042] Unmanned vehicle short-distance trajectory planning algorithm (ASP): This algorithm plans based on the predicted results according to the current movement speed and posture. It predicts whether a collision will occur after reaching the pedestrian position and generates a reasonable movement trajectory. Different from the traditional trajectory planning algorithm, we rely on the predicted pedestrian movement trajectory to expand the scope of trajectory prediction.

[0043] Cloud service allocation algorithm (AIC): In actual traffic conditions, each intersection is only suitable for deploying a small number of MEC devices according to different traffic volumes, which may result in one MEC device serving several driverless cars. At this time, how to allocate task priorities is particularly important. This algorithm is designed to solve the multi-task situation under limited threads, and determine the priority of sending and internal algorithm processing. Implementation of this module: The eNB (total) base station dispatches the V2X terminal to send data, and the base station uniformly allocates air interface resources to ensure that the wireless resources used by each UE do not conflict. Finally, based on the detection of the use of air interface wireless resources in the past period of time, the use of wireless resources in the future period of time can be predicted. The allocation of resources determines whether to send the algorithm results in the current state by judging the request of the lower computer (the main controller in the autonomous driving vehicle). When sending, the vehicle motion trajectory or pedestrian prediction motion trajectory prediction failure information is packaged into a Socket information packet, which is sent to the main controller in the autonomous driving vehicle through the MEC communication module (wireless transmission module). The main controller in the autonomous driving vehicle parses it, and then the main controller makes further decisions autonomously.

[0044] As mentioned above, the edge cloud device that predicts pedestrian behavior based on the perception system predicts the movement trajectory of pedestrians in a certain time axis in the future by modeling the pedestrian behavior characteristics, and uses this information to plan the future movement trajectory of the autonomous vehicle. The pedestrian trajectory feature model is transplanted into the edge cloud device, combined with millimeter wave radar and camera equipment to construct a two-dimensional particle space, and the trajectory of the particle target in the future is predicted based on the real-time pedestrian movement trajectory, so as to plan a shorter and more comfortable autonomous driving vehicle driving path in the future in advance, empowering traffic safety while providing better autonomous driving services or non-autonomous driving assistance services.

[0045] A pedestrian trajectory prediction model (AIP) is provided. The pedestrian motion trajectory feature model obtained through preliminary training can predict the pedestrian trajectory in the future. Combined with the additionally designed judgment algorithm, it can provide accurate and reliable predicted trajectory.

[0046] A sensor data fusion algorithm is provided, which improves the clustering effect of sensor device data clusters, and a two-dimensional particle map modeling method is proposed based on the data set to draw the two-dimensional particle movement trajectory.

[0047] A cloud service task scheduling algorithm (AIC) is proposed, which optimizes the task scheduling mechanism, uses event triggering mechanism, couples multiple MEC units, and implements task priority setting according to the request of the lower computer and the task allocation interrupt mechanism, thereby improving the service efficiency of the MEC device itself.

[0048] The present invention can successfully predict the movement trajectory of pedestrians and publish it to the self-driving car to realize the short-distance trajectory planning of the self-driving car. The prediction method uses the real-time status of the video and data from the real walking process to establish a feature model to perform prediction, which can effectively predict the walking path that has not been walked. Combined with the accurate position information provided by the perception fusion algorithm, the accuracy is greatly improved compared with the accuracy provided by the traditional method based on a single camera.

[0049] Based on V2X technology, an intelligent safety assisted driving system is established. Through wireless communication between vehicles and edge cloud devices, road hazard prediction can be achieved, which can avoid traffic collisions to a great extent and plan better short-distance movement trajectories to empower autonomous driving.

[0050] V2X-based environmental perception can cover the visual blind spots of drivers and sensors, obtain the deep operating status of the target (vehicle braking and steering, historical trajectory, abnormal status, etc.), and accurately analyze map, signal light and traffic sign information. Especially for safety warnings, continuous tracking of the target vehicle status and real-time traffic conditions based on V2X interaction can help drivers or smart driving cars adjust their driving behaviors in time and reduce conflicts in advance before conflicts gradually form. This method does not even need to trigger the conventional ADAS protection mechanism, making the driving process smoother and more energy-efficient.

[0051] On the other hand, MEC devices based on the prediction of pedestrian behavior provide intelligent vehicles with efficient interactive collaboration. The collaboration in the driving process breaks through the previous scene concept based on the perspective of a single vehicle. The higher the level of intelligence of the vehicle, the more comprehensive and accurate the information available for interactive collaboration, and the more significant the improvement in safety and efficiency brought by V2X interactive collaboration.

[0052] Example 2

[0053] On the basis of the aforementioned embodiment 1, in order to optimize data processing efficiency, the road-cloud-coordinated pedestrian prediction-guided autonomous driving path planning system of the present embodiment 2 also includes a screening module, which is used to divide the road into a number of lanes, and use the first pedestrian motion trajectory that intersects with the lane in the upward and downward directions of each lane as a screening condition, and screen out the pedestrian predicted motion trajectory that needs to be judged for reliability from all the pedestrian motion trajectories within a certain time period within the detection range, and output it to the reliability judgment module for reliability judgment. The pedestrian motion trajectory includes the actual motion trajectory of the pedestrian and the predicted motion trajectory of the pedestrian. For example, the road is 15m wide, each lane is 3m wide, and pedestrians can generally walk 0.5m in the predicted time period. The width of each lane can be set to 0.4m, 0.5m, 0.6m, etc.

[0054] It should be pointed out that the examples of the above-mentioned embodiments can be preferably combined with one or more of them according to actual needs, and multiple examples use a set of illustrations of combined technical features, which will not be described one by one here.

[0055] The above descriptions are detailed descriptions and illustrations of the preferred embodiments of the present invention, but these descriptions are not intended to limit the scope of protection required by the present invention. All equivalent changes or modified modifications accomplished under the technical teachings suggested by the present invention should fall within the scope of patent protection covered by the present invention.

Claims

1. An autonomous driving path planning system with road-cloud collaboration and pedestrian prediction guidance, characterized in that: Includes the following: Pedestrian trajectory acquisition module: used to obtain the image behavior information and actual position information of pedestrians, and fuse them to obtain the actual movement trajectory of pedestrians; Behavior prediction module: used to import the actual motion trajectory of multiple frames of pedestrians into the trained pedestrian motion trajectory prediction model to generate the pedestrian predicted motion trajectory; Reliable determination module: used to obtain multiple segments of pedestrian predicted motion trajectories and pedestrian actual motion trajectories in the previous period of the pedestrian predicted motion trajectory, and analyze and calculate their similarities to obtain the final value of the overlap; if the final value of the overlap is greater than or equal to the preset threshold, the prediction is determined to be reasonable, otherwise it is determined to be a prediction failure; Vehicle trajectory planning module: used to import the predicted reasonable pedestrian motion trajectory into the vehicle trajectory planning model, predict whether there will be a collision with the pedestrian based on the current motion speed and posture, and generate the vehicle motion trajectory; Result output module: used to respond to the request of the on-board lower computer and send the prediction failure information of the vehicle motion trajectory or pedestrian prediction motion trajectory to the on-board lower computer through the MEC communication module; The specific operation of the reliable determination module is as follows: obtaining the actual trajectory A of the nth segment, the predicted trajectory B of the Nth segment, and the predicted trajectory C of the N+1th segment of the pedestrian's predicted motion trajectory, analyzing the similarity between the predicted trajectory C of the N+1th segment and the actual trajectory A of the nth segment and the predicted trajectory B of the Nth segment to obtain two coincidences, and summing the two coincidences according to the weight ratio trained by the neural network model to obtain a final coincidence value; if the final coincidence value is greater than or equal to a preset threshold, the prediction is determined to be reasonable; otherwise, the prediction is determined to have failed.

2. The automatic driving path planning system with road-cloud collaboration and pedestrian prediction guidance according to claim 1 is characterized by: The pedestrian trajectory acquisition module specifically operates as follows: by using a calibration algorithm, the millimeter wave radar and the camera group are calibrated to the same coordinate system and associated with the world coordinate system; a clustering algorithm is performed on the millimeter wave radar data to cluster the collected point information of the person into particles to obtain the relative actual position information of the pedestrian; Perform target detection on the camera module at a certain frame rate, use the YOLO algorithm to detect and track targets, and obtain image behavior information of pedestrians; The obtained image behavior information and relative actual position information are fused to establish a two-dimensional particle coordinate map to obtain the actual motion trajectory of the pedestrian with position information.

3. The automatic driving path planning system with road-cloud collaboration and pedestrian prediction guidance according to claim 1 is characterized in that: The specific operation of the behavior prediction module is as follows: first, the position information obtained by fusing multiple frames of pedestrian actual motion trajectories is processed; then, the information under multiple frames is combined and imported into the trained pedestrian motion trajectory prediction model; and then the pedestrian motion trajectory prediction model generates a predicted motion trajectory route and reflects it in a two-dimensional particle coordinate diagram.

4. The automatic driving path planning system with road-cloud collaboration and pedestrian prediction guidance according to claim 1, characterized in that: The preset threshold value of the final value of the overlap degree is 90%.

5. The automatic driving path planning system with road-cloud collaboration and pedestrian prediction guidance according to claim 1, characterized in that: It also includes a screening module, which is used to divide the road into several lanes, and use the first pedestrian movement trajectory that intersects with the lane in the upward and downward directions of each lane as a screening condition. From all the pedestrian movement trajectories within a certain period of time within the detection range, the predicted pedestrian movement trajectory that needs to be judged for reliability is screened out, and output it to the reliable judgment module for reliability judgment, wherein the pedestrian movement trajectory includes the actual pedestrian movement trajectory and the predicted pedestrian movement trajectory.

Images