Parking path planning method and system based on deep learning

Abstract

The invention discloses a parking path planning method and system based on deep learning, relates to the technical field of deep learning, and improves the accuracy and safety of parking. A plurality of candidate parking spaces are matched, a plurality of candidate parking driving paths are set for each candidate parking space, the comprehensive scores of all the candidate parking driving paths are obtained, the comprehensive scores of all the candidate parking driving paths are compared, and the target parking space and the corresponding parking driving path are selected according to the comparison result. A parking driving path is discretized to obtain a plurality of path nodes, whether a demanded vehicle deviates from the parking driving path or not is judged according to the real-time driving angle and the current driving speed of the demanded vehicle when the demanded vehicle passes through each path node, and when the demanded vehicle reaches a target parking space, a multi-mode interaction instruction is generated and executed. And the vehicle is parked until the required vehicle is judged to be parked.

Description

Technical Field

[0001] The present invention relates to the field of deep learning technology, and in particular to a parking path planning method and system based on deep learning. Background Art

[0002] With the acceleration of urbanization and the continuous growth of car ownership, the problem of parking difficulties has become increasingly prominent. Traditional parking methods mainly rely on the driver's experience and skills. It often takes a lot of time and energy to find a suitable parking space and perform parking operations. This problem is particularly prominent in areas with concentrated parking demands such as large parking lots and commercial centers.

[0003] In addition, some existing parking assistance systems have relatively simple functions. Most of them can only provide simple reversing images or distance prompts, and cannot comprehensively and intelligently solve many problems in the parking process. For example, it is not possible to accurately match the driver with a suitable parking space based on the vehicle's real-time position and scene information, and it is difficult to plan an efficient and safe parking path. Moreover, the lack of real-time monitoring and adjustment of the vehicle's driving status during the parking process can easily lead to deviations from the parking path, increasing the difficulty and time cost of parking. For this reason, a parking path planning method and system based on deep learning is provided. Summary of the invention

[0004] In order to solve the above technical problems, the purpose of the present invention is to provide a parking path planning method and system based on deep learning.

[0005] In order to achieve the above object, the present invention provides the following technical solutions: A parking path planning method based on deep learning includes the following steps: Step S1: collecting multi-source scene data of a parking scene by setting a multi-source data collection device, and establishing a visual parking scene model according to the multi-source scene data; Step S2: obtaining the real-time location information and vehicle information of the required vehicle and inputting them into the visual parking scene model, thereby matching a number of candidate parking spaces, and setting a number of candidate parking driving paths for each candidate parking space, obtaining a comprehensive score of each candidate parking driving path, and comparing the comprehensive scores of each candidate parking driving path, selecting a number of candidate parking spaces and corresponding candidate parking driving paths according to the comparison results and sending them to the required vehicle; Step S3: the target parking space and the corresponding parking driving path are selected by the demand vehicle, the parking driving path is discretized to obtain a number of path nodes, and the expected arrival time, the allowed speed range and the pre-steering angle range are respectively set for each path node, and then according to the real-time driving angle and the current driving speed of the demand vehicle when passing each path node, it is determined whether the demand vehicle deviates from the parking driving path; Step S4: When the required vehicle arrives at the target parking space, a multimodal interaction instruction is generated and executed until it is determined that the required vehicle has completed parking.

[0006] Furthermore, the parking scene is installed with several multi-source data acquisition devices and an MEC server. The multi-source data acquisition devices include cameras, lidars, ultrasonic sensors and millimeter-wave radars, and parking space detection devices are set for each parking space. The parking space detection devices include GPS positioning devices, cameras and wireless communication devices.

[0007] Furthermore, the process of collecting multi-source scene data of parking scenes includes: A scene data acquisition cycle is set for each multi-source data acquisition device. Whenever a data acquisition cycle begins, the camera in each multi-source data acquisition device is used to capture optical image data of the parking scene, the laser radar generates three-dimensional point cloud data of the parking scene, the ultrasonic sensor detects the distance information of obstacles in the parking scene, and the millimeter wave radar obtains the motion trajectory of the dynamic target in real time.

[0008] Furthermore, the process of establishing the visual parking scene model includes: A 3D parking scene model is established based on the optical image data of the parking scene, and the 3D parking scene model and the 3D point cloud data are split into a number of data display cubes of the same size, and then the data display cubes are matched with each other, and the 3D parking scene model and the 3D point cloud data are fused according to the matching results to obtain a visualized parking scene model; The distance information of obstacles, the motion trajectory of dynamic targets, and the location information of parking spaces are mapped to the visual parking scene model, and a number of scene semantic objects are segmented in the visual parking scene model through a deep learning algorithm.

[0009] Furthermore, the process of selecting the candidate parking space includes: The required vehicle is equipped with a GPS positioning device, an inertial measurement unit and a wireless communication device. When the required vehicle enters the parking scene, the MEC server automatically communicates with the wireless communication device of the required vehicle, and automatically uploads the vehicle information, real-time location information and real-time heading angle of the required vehicle through the GPS positioning device and the inertial measurement unit; According to the vehicle information, real-time location information and real-time heading angle of the required vehicle, a corresponding scene semantic object is automatically generated in the visual parking scene model, and various information of the required vehicle is marked on the scene semantic object; According to the currently idle parking spaces and the corresponding length and width values ​​in the visual parking scene model, as well as the size information of the required vehicle, a greedy algorithm is used to select parking spaces that are larger than the length and width of the required vehicle, and the selected parking spaces are recorded as candidate parking spaces.

[0010] Furthermore, the process of setting a pending parking driving route and obtaining a corresponding comprehensive score includes: Based on the spatial position of the candidate parking space in the visualized parking scene model and the real-time position information of the required vehicle, a number of pending parking driving paths are preset for each candidate parking space; A unit driving speed is set, and then each pending parking driving path and the parking driving path being executed in the parking scene are simulated according to the unit driving speed; Selecting a pending parking driving path without collision according to the driving simulation result and recording it as a candidate parking driving path for each candidate parking space; Obtain the comprehensive score of each candidate parking path for each candidate parking space, The candidate parking paths of all candidate parking spaces are sorted in descending order according to their comprehensive scores, and then the MEC server selects the ten candidate parking spaces with the highest comprehensive scores and the corresponding candidate parking paths and sends them to the required vehicle.

[0011] Furthermore, the calculation formula of the comprehensive score is: ; in , , Calculate the weights for efficiency, and , P and T represent the number of obstacles appearing in the candidate parking path and the estimated time required for driving at a unit speed, and n, θ, and L represent the number of turns, the average turning angle of each turn, and the total length of the candidate parking path, respectively.

[0012] Furthermore, the process of determining whether the required vehicle deviates from the parking path includes: The demand vehicle selects the target parking space and the corresponding parking driving path from the candidate parking spaces and the corresponding candidate parking driving paths sent by the MEC server; Set a unit detection distance, divide the parking driving path into multiple path nodes according to the unit detection distance, and set the expected arrival time, allowed speed range and pre-steering angle range for each path node. It should be noted that the allowed speed range and pre-steering angle range of different path nodes are different, and there are path nodes without pre-steering angle range; When the demand vehicle passes through each path node in sequence according to the parking driving path, the MEC server synchronously updates the visual parking scene model. If it is determined that there are new obstacles or deviations from the predicted path at the current untraveled path node of any demand vehicle, the parking driving path is planned for the corresponding demand vehicle, otherwise no operation is performed; At the same time, when the required vehicle passes each path node, its real-time driving angle and current driving speed are obtained through the inertial measurement unit on the required vehicle, so as to judge whether the current driving speed of the required vehicle at each path node is within the allowed speed range, and whether the real-time driving angle at the corner of the path node is within the pre-steering angle range; If it is determined that either the real-time driving angle or the current driving speed is not within the corresponding range, the corresponding vehicle is judged to be driving abnormally, and the corresponding speed or steering angle repair decision is generated through the PID control algorithm until it is determined that both the real-time driving angle and the current driving speed are within the corresponding range, otherwise no operation is performed; The above process of determining whether the required vehicle has deviated from the parking driving path is repeated until the required vehicle reaches the target parking space.

[0013] Furthermore, when the required vehicle reaches the target parking space, the scene semantic objects corresponding to the required vehicle and the target parking space are superimposed in the visual parking scene model, and then a multimodal interactive instruction is generated to provide parking guidance for the required vehicle; When it is determined that the distance between both sides of the required vehicle and the edge line of the target parking space is between 0.2 and 0.5m, and the extension angle between the head or tail of the required vehicle and the tail line of the target parking space does not exceed plus or minus three to five degrees, and the vertical distance is between 0.3 and 0.5m, the required vehicle is judged to have completed parking, otherwise the multimodal interaction command will continue to be executed.

[0014] A parking path planning system based on deep learning, including a scene data acquisition module, a path planning module, a driving supervision module and a parking guidance module; The scene data acquisition module is used to collect multi-source scene data of the parking scene by setting a multi-source data acquisition device, and establish a visual parking scene model according to the multi-source scene data; The path planning module is used to obtain the real-time location information and vehicle information of the required vehicle, and input them into the visual parking scene model to match the candidate parking spaces, set a number of candidate parking driving paths for each candidate parking space, and obtain the comprehensive score of each candidate parking driving path; The driving supervision module is used to select a target parking space and a corresponding parking driving path, discretize the parking driving path to obtain a number of path nodes, and set an expected arrival time, an allowable speed range, and a pre-steering angle range for each path node, and then judge whether the required vehicle deviates from the parking driving path according to the real-time driving angle and current driving speed of the required vehicle when passing each path node; The parking guidance module is used to generate and execute multimodal interactive instructions, and to determine in real time whether the required vehicle has completed parking.

[0015] Compared with the prior art, the present invention has the following beneficial effects: The present invention sets the expected arrival time, allowed speed range and pre-steering angle range of each path node, and determines whether the path is deviated according to the real-time driving angle and speed of the vehicle. It can timely detect and correct the deviation of the vehicle during parking, and ensure that the vehicle is parked accurately according to the planned path, thereby improving the accuracy and safety of parking. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in the present invention.

[0017] Figure 1 The figure is a flow chart of the method of the present invention. DETAILED DESCRIPTION

[0018] To make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be described in detail below. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other implementation methods obtained by ordinary technicians in this field without creative work belong to the scope of protection of the present invention.

[0019] like Figure 1 As shown, a parking path planning method based on deep learning includes the following steps: Step S1: collecting multi-source scene data of a parking scene by setting a multi-source data collection device, and establishing a visual parking scene model according to the multi-source scene data; Step S2: obtaining the real-time location information and vehicle information of the required vehicle and inputting them into the visual parking scene model, thereby matching a number of candidate parking spaces, and setting a number of candidate parking driving paths for each candidate parking space, obtaining a comprehensive score of each candidate parking driving path, and comparing the comprehensive scores of each candidate parking driving path, selecting a number of candidate parking spaces and corresponding candidate parking driving paths according to the comparison results and sending them to the required vehicle; Step S3: the target parking space and the corresponding parking driving path are selected by the demand vehicle, the parking driving path is discretized to obtain a number of path nodes, and the expected arrival time, the allowed speed range and the pre-steering angle range are respectively set for each path node, and then according to the real-time driving angle and the current driving speed of the demand vehicle when passing each path node, it is determined whether the demand vehicle deviates from the parking driving path; Step S4: When the required vehicle arrives at the target parking space, a multimodal interaction instruction is generated and executed until it is determined that the required vehicle has completed parking.

[0020] The step S1 is implemented by the following process: Several multi-source data acquisition devices and an MEC server are installed in the parking scene. The multi-source data acquisition devices include cameras, laser radars, ultrasonic sensors, and millimeter-wave radars. A parking space detection device is set for each parking space. The parking space detection device includes a GPS positioning device, a camera, and a wireless communication device. The MEC server has a built-in local area network, and the signal coverage includes the entire parking scene; It should be noted that the data collection ranges of the various multi-source data collection devices are the same, and the data collection ranges of the multi-source data collection devices at adjacent spatial positions overlap; Setting a scene data collection cycle for each multi-source data collection device, wherein the duration of the scene data collection cycle is generally 50 to 100 ms; Whenever a data collection cycle begins, the camera in each multi-source data collection device is used to capture optical image data of the parking scene, the laser radar generates three-dimensional point cloud data of the parking scene, the ultrasonic sensor detects the distance information of obstacles in the parking scene, and the millimeter wave radar obtains the motion trajectory of the dynamic target in real time; A scene fusion base is set up according to the area occupied by the parking scene, and a 3D parking scene model is established according to the optical image data of the parking scene. Then, the 3D parking scene model and the 3D point cloud data are first mapped to the scene fusion base. Since the 3D parking scene model and the 3D point cloud data are both generated corresponding to the parking scene, there is a one-to-one correspondence between the scene information contained in the two. Then, the 3D parking scene model and the 3D point cloud data are split into a number of data display cubes of the same size, and then the data display cubes are matched with each other. According to the matching results, the 3D parking scene model and the 3D point cloud data are fused to obtain a visualized parking scene model. The distance information of obstacles, the motion trajectory of dynamic targets and the location information of parking spaces are mapped to the visual parking scene model, and several scene semantic objects are segmented in the visual parking scene model through a deep learning algorithm; The scene semantic object represents a local scene model with semantic marks in the visualized parking scene model, and the semantic marks include obstacle names, dynamic target marks, and parking space marks; It should be noted that, for the scene semantic object corresponding to the parking space, whether the parking space is currently idle and the length and width values ​​are marked thereon.

[0021] The step S2 is implemented by the following process: The required vehicle is equipped with a GPS positioning device, an inertial measurement unit and a wireless communication device. When the required vehicle enters the parking scene, the MEC server automatically communicates with the wireless communication device of the required vehicle, and automatically uploads the vehicle information, real-time location information and real-time heading angle of the required vehicle through the GPS positioning device and the inertial measurement unit; The vehicle information includes vehicle size, minimum turning radius and current load status; According to the vehicle information, real-time location information and real-time heading angle of the required vehicle, a corresponding scene semantic object is automatically generated in the visual parking scene model, and various information of the required vehicle is marked on the scene semantic object; According to the currently idle parking spaces and the corresponding length and width values ​​in the visual parking scene model, combined with the size information of the required vehicle, a greedy algorithm is used to select parking spaces that are larger than the length and width of the required vehicle, and the selected parking spaces are recorded as candidate parking spaces; Based on the spatial position of the candidate parking space in the visualized parking scene model and the real-time position information of the required vehicle, a number of pending parking driving paths are preset for each candidate parking space; A unit driving speed is set, and then each pending parking driving path and the parking driving path being executed in the parking scene are simulated according to the unit driving speed; Selecting a pending parking driving path without collision according to the driving simulation result and recording it as a candidate parking driving path for each candidate parking space; The comprehensive score of each candidate parking driving path of each candidate parking space is calculated, and the calculation formula of the comprehensive score is: ; in , , Calculate the weights for efficiency, and , P and T represent the number of obstacles in the candidate parking path and the estimated time required for driving at a unit speed, n, θ and L represent the number of turns, the average turning angle of each turn and the total length of the candidate parking path respectively; The candidate parking paths of all candidate parking spaces are sorted in descending order according to their comprehensive scores, and then the MEC server selects the ten candidate parking spaces with the highest comprehensive scores and the corresponding candidate parking paths and sends them to the required vehicle.

[0022] The step S3 is implemented by the following process: The demand vehicle selects the target parking space and the corresponding parking driving path from the candidate parking spaces and the corresponding candidate parking driving paths sent by the MEC server; Set a unit detection distance, divide the parking driving path into multiple path nodes according to the unit detection distance, and set the expected arrival time, allowed speed range and pre-steering angle range for each path node. It should be noted that the allowed speed range and pre-steering angle range of different path nodes are different, and there are path nodes without pre-steering angle range; When the demand vehicle passes through each path node in sequence according to the parking driving path, the MEC server synchronously updates the visual parking scene model. If it is determined that there are new obstacles or deviations from the predicted path at the current untraveled path node of any demand vehicle, the parking driving path is planned for the corresponding demand vehicle, otherwise no operation is performed; At the same time, when the required vehicle passes each path node, its real-time driving angle and current driving speed are obtained through the inertial measurement unit on the required vehicle, so as to judge whether the current driving speed of the required vehicle at each path node is within the allowed speed range, and whether the real-time driving angle at the corner of the path node is within the pre-steering angle range; If it is determined that either the real-time driving angle or the current driving speed is not within the corresponding range, the corresponding vehicle is judged to be driving abnormally, and the corresponding speed or steering angle repair decision is generated through the PID control algorithm until it is determined that both the real-time driving angle and the current driving speed are within the corresponding range, otherwise no operation is performed; The above process of determining whether the required vehicle has deviated from the parking driving path is repeated until the required vehicle reaches the target parking space.

[0023] The step S4 is implemented by the following process: When the required vehicle reaches the target parking space, the scene semantic objects corresponding to the required vehicle and the target parking space are superimposed in the visual parking scene model, and then multimodal interactive instructions are generated to provide parking guidance for the required vehicle; The multimodal interaction instructions include: Visual prompts: The camera in the parking space detection device of the target parking space obtains the image data of the required vehicle relative to the target parking space in real time, and displays its relative position to the target parking space through the on-board display screen of the required vehicle, and marks the steering wheel steering angle suggestion; Voice prompts: broadcast real-time operation instructions, such as "nudge left 15 degrees" or "slow down to 2km / h"; Automatic control command: If the vehicle is equipped with a wire-controlled chassis, the steering motor and drive motor are directly controlled to perform the parking action, otherwise the command is automatically skipped; When it is determined that the distance between both sides of the required vehicle and the edge line of the target parking space is between 0.2 and 0.5m, and the extension angle between the head or tail of the required vehicle and the tail line of the target parking space does not exceed plus or minus three to five degrees, and the vertical distance is between 0.3 and 0.5m, then the required vehicle is judged to have completed parking, otherwise the multimodal interaction command will continue to be executed.

[0024] The present invention also discloses a parking path planning system based on deep learning, including a scene data acquisition module, a path planning module, a driving supervision module and a parking guidance module; The scene data acquisition module is used to collect multi-source scene data of the parking scene by setting a multi-source data acquisition device, and establish a visual parking scene model according to the multi-source scene data; The path planning module is used to obtain the real-time location information and vehicle information of the required vehicle, and input them into the visual parking scene model to match the candidate parking spaces, set a number of candidate parking driving paths for each candidate parking space, and obtain the comprehensive score of each candidate parking driving path; The driving supervision module is used to select a target parking space and a corresponding parking driving path, discretize the parking driving path to obtain a number of path nodes, and set an expected arrival time, an allowable speed range, and a pre-steering angle range for each path node, and then judge whether the required vehicle deviates from the parking driving path according to the real-time driving angle and current driving speed of the required vehicle when passing each path node; The parking guidance module is used to generate and execute multimodal interactive instructions, and to determine in real time whether the required vehicle has completed parking.

[0025] The above contents are only used to illustrate the technical method of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical method of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical method of the present invention.

Claims

1. A parking path planning method based on deep learning, characterized in that: The following steps are involved: Step S1: collecting multi-source scene data of a parking scene by setting a multi-source data collection device, and establishing a visual parking scene model according to the multi-source scene data; Step S2: obtaining the real-time location information and vehicle information of the required vehicle and inputting them into the visual parking scene model, thereby matching a number of candidate parking spaces, and setting a number of candidate parking driving paths for each candidate parking space, obtaining a comprehensive score of each candidate parking driving path, and comparing the comprehensive scores of each candidate parking driving path, selecting a number of candidate parking spaces and corresponding candidate parking driving paths according to the comparison results and sending them to the required vehicle; Step S3: the target parking space and the corresponding parking driving path are selected by the demand vehicle, the parking driving path is discretized to obtain a number of path nodes, and the expected arrival time, the allowed speed range and the pre-steering angle range are respectively set for each path node, and then according to the real-time driving angle and the current driving speed of the demand vehicle when passing each path node, it is determined whether the demand vehicle deviates from the parking driving path; Step S4: When the required vehicle arrives at the target parking space, a multimodal interaction instruction is generated and executed until it is determined that the required vehicle has completed parking.

2. The parking path planning method based on deep learning according to claim 1, characterized in that: The parking scene is equipped with several multi-source data acquisition devices and a MEC server.

3. The parking path planning method based on deep learning according to claim 2, characterized in that: The process of collecting multi-source scene data for parking scenes includes: A scene data acquisition cycle is set for each multi-source data acquisition device. When a data acquisition cycle starts, optical image data, three-dimensional point cloud data, distance information of obstacles, motion trajectory of dynamic targets and location information of parking spaces of the parking scene are collected.

4. The parking path planning method based on deep learning according to claim 3, characterized in that: The process of establishing the visual parking scene model includes: A 3D parking scene model is established based on the optical image data of the parking scene, and the 3D parking scene model and the 3D point cloud data are split into a number of data display cubes of the same size, and then the data display cubes are matched with each other, and the 3D parking scene model and the 3D point cloud data are fused according to the matching results to obtain a visualized parking scene model; The distance information of obstacles, the motion trajectory of dynamic targets, and the location information of parking spaces are mapped to the visual parking scene model, and a number of scene semantic objects are segmented in the visual parking scene model through a deep learning algorithm.

5. The parking path planning method based on deep learning according to claim 4, characterized in that: The process of selecting the candidate parking space includes: The required vehicle is equipped with a GPS positioning device, an inertial measurement unit and a wireless communication device. When the required vehicle enters the parking scene, the MEC server automatically communicates with the wireless communication device of the required vehicle, and automatically uploads the vehicle information, real-time location information and real-time heading angle of the required vehicle through the GPS positioning device and the inertial measurement unit; According to the vehicle information, real-time location information and real-time heading angle of the required vehicle, a scene semantic object is automatically generated in the visual parking scene model, and various information of the required vehicle is marked on the scene semantic object; According to the currently idle parking spaces and the corresponding length and width values ​​in the visual parking scene model, combined with the size information of the required vehicle, a greedy algorithm is used to select parking spaces that are larger than the length and width of the required vehicle, and the selected parking spaces are recorded as candidate parking spaces.

6. The parking path planning method based on deep learning according to claim 5, characterized in that: The process of setting a pending parking driving route and obtaining the corresponding comprehensive score includes: Based on the spatial position of the candidate parking space in the visualized parking scene model and the real-time position information of the required vehicle, a number of pending parking driving paths are preset for each candidate parking space; A unit driving speed is set, and then each pending parking driving path and the parking driving path being executed in the parking scene are simulated according to the unit driving speed; Selecting a pending parking driving path without collision according to the driving simulation result and recording it as a candidate parking driving path for each candidate parking space; The comprehensive scores of the candidate parking paths of the candidate parking spaces are obtained, and the candidate parking paths of all the candidate parking spaces are sorted from high to low according to the comprehensive scores, and the ten candidate parking spaces with the highest comprehensive scores and the corresponding candidate parking paths are selected and sent to the demand vehicle.

7. The parking path planning method based on deep learning according to claim 6, characterized in that: The process of determining whether the required vehicle deviates from the parking path includes: The vehicle in need selects the target parking space and the corresponding parking driving path, sets the unit detection distance, divides the parking driving path into multiple path nodes according to the unit detection distance, and sets the expected arrival time, allowed speed range and pre-steering angle range for each path node; When the demand vehicle passes through each path node in sequence according to the parking driving path, the MEC server synchronously updates the visual parking scene model. If it is determined that there are new obstacles or deviations from the predicted path at the current untraveled path node of any demand vehicle, the parking driving path is planned for the corresponding demand vehicle, otherwise no operation is performed; At the same time, when the required vehicle passes each path node, its real-time driving angle and current driving speed are obtained through the inertial measurement unit on the required vehicle, so as to judge whether the current driving speed of the required vehicle at each path node is within the allowed speed range, and whether the real-time driving angle at the corner of the path node is within the pre-steering angle range; If it is determined that either the real-time driving angle or the current driving speed is not within the corresponding range, the corresponding vehicle is judged to be driving abnormally, otherwise no operation is performed.

8. The parking path planning method based on deep learning according to claim 7, characterized in that: When the required vehicle reaches the target parking space, the scene semantic objects corresponding to the required vehicle and the target parking space are superimposed in the visual parking scene model, and then multimodal interactive instructions are generated to guide the required vehicle in parking until it is determined that the required vehicle has completed parking.

9. A parking path planning system based on deep learning, used to implement a parking path planning method based on deep learning according to any one of claims 1 to 8, characterized in that: It includes scene data collection module, path planning module, driving supervision module and parking guidance module; The scene data acquisition module is used to collect multi-source scene data of the parking scene by setting a multi-source data acquisition device, and establish a visual parking scene model according to the multi-source scene data; The path planning module is used to obtain the real-time location information and vehicle information of the required vehicle, and input them into the visual parking scene model to match the candidate parking spaces, set a number of candidate parking driving paths for each candidate parking space, and obtain the comprehensive score of each candidate parking driving path; The driving supervision module is used to select a target parking space and a corresponding parking driving path, discretize the parking driving path to obtain a number of path nodes, and set an expected arrival time, an allowable speed range, and a pre-steering angle range for each path node, and then judge whether the required vehicle deviates from the parking driving path according to the real-time driving angle and current driving speed of the required vehicle when passing each path node; The parking guidance module is used to generate and execute multimodal interactive instructions, and to determine in real time whether the required vehicle has completed parking.

Images