Automatic parking obstacle avoidance method based on vehicle-mounted image data

Abstract

The invention discloses an automatic parking obstacle avoidance method based on vehicle-mounted image data, relates to the technical field of automatic parking, and effectively improves the safety and reliability of automatic parking. A dynamic parking scene model with a vehicle as the center is established through a multi-view scene image, a plurality of scene object models are divided in the dynamic parking scene model, a scene semantic tag is set for each scene object model, and the spatial position of the scene object model in the next scene model updating period is obtained. According to the spatial position of the scene object model in the next scene model updating period, whether collision occurs in the collision-free parking path or not is judged, a plurality of bypassing collision-free parking paths are generated according to the judgment result, and the real-time risk value of each bypassing collision-free parking path is obtained; and selecting the bypassing collision-free parking path with the minimum real-time risk value as a new collision-free parking path, and executing the new collision-free parking path.

Description

Technical Field

[0001] The present invention relates to the technical field of automatic parking, and specifically to an automatic parking obstacle avoidance method based on vehicle-mounted image data. Background Art

[0002] In modern society, with the sharp increase in the number of motor vehicles, the problem of difficult parking has become increasingly prominent. To improve the convenience and safety of parking, automatic parking technology has emerged. Automatic parking technology can help drivers automatically find suitable parking spaces and control the vehicle to complete the parking operation, greatly reducing the operation burden of drivers during the parking process.

[0003] However, there are still some problems in the actual application of existing automatic parking technologies. On the one hand, due to factors such as the installation position, angle, and hardware parameters of vehicle-mounted cameras, the image data collected by different cameras may have perspective differences and distortions, resulting in inaccurate multi-perspective scene images after stitching, making it difficult to truly reflect the actual parking scene, thus affecting the accuracy of subsequent parking path planning.

[0004] On the other hand, during the parking process, the objects in the surrounding environment (such as other vehicles, pedestrians, obstacles, etc.) are dynamically changing. Existing parking path planning methods often cannot effectively respond to these dynamic changes in a timely manner. When encountering obstacles, it may not be able to quickly plan a reasonable detour path, or the planned detour path has a high risk and is prone to collision accidents. Therefore, an automatic parking obstacle avoidance method based on vehicle-mounted image data is provided. Summary of the Invention

[0005] In order to solve the above technical problems, the purpose of the present invention is to provide an automatic parking obstacle avoidance method based on vehicle-mounted image data.

[0006] To achieve the above purpose, the present invention provides the following technical solutions:

[0007] An automatic parking obstacle avoidance method based on vehicle-mounted image data, comprising the following steps:

[0008] Step S1: Establish a spatial calibration board, map the parking scene image data collected by the vehicle-mounted surround-view camera and the front-view fish-eye camera onto the spatial calibration board, mark several corner points therein, match the corner points of different parking scene image data, record some corner points as inliers according to the matching results, and then adjust the hardware parameters of the vehicle-mounted surround-view camera according to the positions of the inliers. Then, splice the parking scene image data collected by the adjusted vehicle-mounted surround-view camera and the front-view fish-eye camera to obtain a multi-perspective scene image;

[0009] Step S2: Establish a vehicle-centered dynamic parking scenario model based on multi-perspective scenario images, divide several scenario object models in the dynamic parking scenario model, and set scenario semantic labels for each scenario object model;

[0010] Step S3: Select a target parking position in the dynamic parking scenario model, generate a corresponding collision-free parking path and set the scenario model update period, and then obtain the spatial positions of the scenario object models in the next scenario model update period;

[0011] Step S4: According to the spatial positions of the scenario object models in the next scenario model update period, determine whether the collision-free parking path will collide, generate multiple detour collision-free parking paths according to the judgment result, obtain the real-time risk values of each detour collision-free parking path, and then select the detour collision-free parking path with the smallest real-time risk value as the new collision-free parking path and execute it.

[0012] Further, the process of annotating corner points in the parking scenario image data includes:

[0013] Deploy four on-vehicle surround-view cameras and one front-view fish-eye camera on the vehicle. The on-vehicle surround-view cameras are located in the front, rear, left, right, and front wide-angle areas of the vehicle, and each on-vehicle surround-view camera and the front-view fish-eye camera are distributed according to spatial positions, and there is an overlapping part in the shooting ranges of cameras at adjacent spatial positions;

[0014] When the vehicle generates a parking request, first make each on-vehicle surround-view camera and one front-view fish-eye camera simultaneously capture 10 to 20 pieces of parking scenario image data from different angles and establish a spatial calibration board;

[0015] The spatial calibration board is composed of several spatial squares of the same size, and the volume of each spatial square is equal to a spatial cube composed of six pixel planes;

[0016] Map each piece of parking scenario image data onto the spatial calibration board according to the camera and the shooting angle, and mark several corner points in each piece of parking scenario image data through OpenCV.

[0017] Further, the process of matching corner points of different parking scenario image data includes:

[0018] Taking the parking scenario image data captured by the front-view fish-eye camera as the reference image data, first match the corner points in the parking scenario image data captured by the front-view fish-eye camera and the on-vehicle surround-view cameras at its adjacent positions;

[0019] If it is determined that two corner points correspond to the same position, it is determined that the two corner points match; otherwise, it is determined that they do not match, and the mutually matching corner points are grouped, with each group containing several corner points.

[0020] Further, the process of adjusting the hardware parameters of each vehicle-mounted surround-view camera includes:

[0021] Obtain the model parameters of each group of corner points, where the model parameters are used to represent the coordinate position parameters of the corner points of different parking scenario image data mapped onto the image data of another parking scenario according to the matching relationship.

[0022] Set the planar distance threshold and the spatial distance threshold, and according to the model parameters, obtain the coordinate positions of the corner points corresponding to any vehicle-mounted surround-view camera in the same group mapped onto the parking scenario image data collected by the front-view fisheye camera, and then obtain the coordinate distance between the corresponding two corner points in the same parking scenario image data.

[0023] If the coordinate distance is greater than or equal to the planar distance threshold, no operation is performed.

[0024] If the coordinate distance is less than the planar distance threshold, the corresponding two corner points are marked as inliers.

[0025] Furthermore, according to the inlier correspondence relationship between the image data of each parking scenario, the internal parameter matrix and the external parameter matrix of each camera are obtained by the Zhang Zhengyou calibration method.

[0026] Obtain the world coordinate positions of each inlier through the internal parameter matrix and the external parameter matrix of each camera, where the world coordinate positions represent the actual spatial positions of each inlier according to the hardware parameters of the camera.

[0027] Map the world coordinate positions of the inliers corresponding to each vehicle-mounted surround-view camera onto the spatial calibration board, and judge the spatial distance between the coordinate positions of each inlier on the spatial calibration board and the corresponding world coordinate positions. If the spatial distance is greater than or equal to the spatial distance threshold, adjust the hardware parameters of the corresponding vehicle-mounted surround-view camera until the spatial distance is less than the spatial distance threshold.

[0028] Further, the process of obtaining the multi-view scene images includes:

[0029] After the hardware parameters of each vehicle-mounted surround-view camera adjacent to the front-view fisheye camera are adjusted, use the same method to adjust the hardware parameters of the remaining two vehicle-mounted surround-view cameras according to the parking scenario image data collected by each vehicle-mounted surround-view camera adjacent to the front-view fisheye camera.

[0030] After the hardware parameters of all on-vehicle surround-view cameras are adjusted, each camera continuously collects parking scene image data within its shooting range. According to the inlier correspondence between the on-vehicle surround-view cameras and the front fisheye camera, the parking scene image data collected at each same time node is stitched together to obtain a multi-view scene image.

[0031] Further, the process of dividing a number of scene object models in the dynamic parking scene model includes:

[0032] Based on the multi-view scene image, a dynamic parking scene model of the parking scene is established, and a three-dimensional coordinate system is established with the vehicle as the origin. Then, the dynamic parking scene model is mapped onto the three-dimensional coordinate system;

[0033] The dynamic parking scene model is divided into a number of model voxels of the same size. The model voxels are cubic, and each face has an independent planar pixel value;

[0034] Starting from the model voxels at the eight vertex positions of the dynamic parking scene model simultaneously, the pixel averaging is performed on each model voxel and its adjacent model voxels, that is, the pixels on the same face of the model voxel and the adjacent model voxels are added and then the average value is taken, and the average value is re-assigned to the corresponding face. No operation is performed on the positions where the model voxel has no adjacent face;

[0035] Repeat the above model voxel pixel averaging operation until the model voxels in eight directions converge to the same position for model voxel pixel averaging;

[0036] When the model voxel pixel averaging of the dynamic parking scene model is completed, the Gaussian mixture clustering algorithm is used to cluster the pixel of each voxel. Then, a number of scene object models are divided in the dynamic parking scene model, and scene semantic labels are set for each scene object model;

[0037] Set the scene model update period, randomly calibrate n motion feature points for each scene object model. Then, whenever a scene model update period ends, detect whether the motion feature points have displacement. n is a natural number greater than 0;

[0038] If it is detected that the motion feature points have displacement in three consecutive scene model update periods, the corresponding scene object model is recorded as a dynamic object scene model, otherwise it is recorded as a static scene object model.

[0039] Further, the process of obtaining the spatial position of the scene object model in the next scene model update period includes:

[0040] Select a target parking position in the dynamic parking scenario model, and then use the improved rapidly-exploring random tree algorithm. Mark the current position of the vehicle as the starting point and the target parking position as the ending point in the dynamic parking scenario model, and each dynamic object scenario model and static scenario object model as obstacles, and traverse a collision-free parking path in the dynamic parking scenario model;

[0041] At the beginning of each scenario model update cycle, quantify the prediction uncertainty according to Gaussian process regression, and predict the spatial positions of each dynamic object scenario model in the next scenario model update cycle;

[0042] Map the predicted spatial coordinates of the motion feature points of each dynamic object scenario model to the displacement positions of each dynamic object scenario model in the dynamic parking scenario model, and mark the displacement direction.

[0043] Furthermore, the generation process of the detouring collision-free parking path includes:

[0044] Whenever the dynamic parking scenario model updates the displacement positions of each dynamic object scenario model in the next scenario model update cycle, determine whether the vehicle will collide in the next scenario model update cycle along the collision-free parking path;

[0045] If it is determined that there is no collision, do nothing;

[0046] If it is determined that there is a collision, regenerate multiple detouring collision-free parking paths with the current position of the vehicle as the starting point and the target parking position as the ending point, and divide each detouring collision-free parking path into several driving path nodes according to the current driving speed of the vehicle and the time length of the scenario model update cycle, and obtain the real-time risk values of each detouring collision-free parking path.

[0047] Furthermore, the obtaining process of the real-time risk values of each detouring collision-free parking path includes:

[0048] Set a real-time risk threshold, compare the real-time risk value of the detouring collision-free parking path with the minimum real-time risk value. If the minimum real-time risk value is less than or equal to the real-time risk threshold, replace the original dynamic collision-free parking path with the corresponding detouring collision-free parking path;

[0049] If the minimum real-time risk value is greater than the real-time risk threshold, there is no executable parking path currently;

[0050] Repeat the above determination process every time a scenario model update cycle passes until the vehicle reaches the target parking position.

[0051] Compared with the prior art, the beneficial effects of the present invention are:

[0052] 1. The present invention maps the parking scene image data collected by the vehicle surround-view camera and the front fisheye camera onto a spatial calibration board, marks the corner points for matching, and adjusts the hardware parameters of the vehicle surround-view camera according to the matching results, effectively eliminating the perspective differences and distortions between the images collected by different cameras, making the multi-perspective scene image after stitching more accurately reflect the actual parking scene, and providing a more reliable data basis for subsequent parking path planning.

[0053] 2. According to the spatial position of the scene object model in the next scene model update cycle, it is judged whether the collision-free parking path will collide, and multiple detour collision-free parking paths are generated. The detour collision-free parking path with the minimum real-time risk value is selected as the new execution path, realizing the dynamic planning of the parking path. At the same time, it can timely respond to the dynamic changes in the surrounding environment during the parking process, effectively avoid the occurrence of collision accidents, and improve the safety and reliability of automatic parking. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings required in the embodiments. Obviously, the drawings described below are only some embodiments recorded in the present invention.

[0055] Figure 1 It is a flowchart of the method of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0056] To make the objectives, technical solutions, and advantages of the present invention clearer, the technical solutions of the present invention will be described in detail below. Obviously, the described embodiments are only some embodiments of the present invention, rather than all embodiments. All other embodiments obtained by those of ordinary skill in the art without creative efforts based on the embodiments of the present invention belong to the scope protected by the present invention.

[0057] As Figure 1 shown, the automatic parking obstacle avoidance method based on vehicle-mounted image data includes the following steps:

[0058] Step S1: Establish a spatial calibration board, map the parking scene image data collected by the vehicle surround-view camera and the front fisheye camera onto the spatial calibration board, mark several corner points therein, match the corner points of different parking scene image data, record some corner points as inliers according to the matching results, and then adjust the hardware parameters of the vehicle surround-view camera according to the inlier positions. Then, splice the parking scene image data collected by the adjusted vehicle surround-view camera and the front fisheye camera to obtain a multi-perspective scene image;

[0059] Step S2: Establish a vehicle-centered dynamic parking scenario model based on multi-view scenario images, divide several scenario object models in the dynamic parking scenario model, and set scenario semantic labels for each scenario object model;

[0060] Step S3: Select a target parking position in the dynamic parking scenario model, generate a corresponding collision-free parking path and set the scenario model update period, and then obtain the spatial positions of the scenario object models in the next scenario model update period;

[0061] Step S4: According to the spatial positions of the scenario object models in the next scenario model update period, determine whether the collision-free parking path will collide, generate multiple detour collision-free parking paths according to the judgment result, obtain the real-time risk values of each detour collision-free parking path, and then select the detour collision-free parking path with the smallest real-time risk value as the new collision-free parking path and execute it.

[0062] Further, the above Step S1 is implemented through the following process:

[0063] Deploy four on-vehicle surround-view cameras and one front-view fisheye camera on the vehicle. The on-vehicle surround-view cameras are located in the front, rear, left, right and front wide-angle areas of the vehicle, and each on-vehicle surround-view camera and the front-view fisheye camera are distributed according to spatial positions, and there is an overlapping part in the shooting ranges of cameras at adjacent spatial positions;

[0064] When the vehicle generates a parking request, first make each on-vehicle surround-view camera and one front-view fisheye camera simultaneously capture 10 to 20 pieces of parking scenario image data from different angles, and establish a spatial calibration board;

[0065] The spatial calibration board is composed of several spatial squares of the same size, and the volume of each spatial square is equal to a spatial cube composed of six pixel planes;

[0066] Map each piece of parking scenario image data onto the spatial calibration board according to the camera and the shooting angle, and mark several corner points in each piece of parking scenario image data through OpenCV;

[0067] Since there is an overlapping part in the shooting ranges of adjacent cameras, there must be the same part between the parking scenario image data captured by adjacent cameras. Therefore, taking the parking scenario image data captured by the front-view fisheye camera as the reference image data, first match the corner points in the parking scenario image data captured by the front-view fisheye camera and the on-vehicle surround-view cameras at its adjacent positions;

[0068] If it is determined that two corner points correspond to the same position, it is determined that the two corner points match; otherwise, it is determined that they do not match. At the same time, since each vehicle-mounted surround-view camera and the front-view fisheye camera capture multiple parking scene image data from different angles, the mutually matching corner points can be grouped, and each group contains several corner points;

[0069] Obtain the model parameters of each group of corner points. The model parameters are used to represent the coordinate position parameters of the corner points of different parking scene image data mapped onto another parking scene image data according to the matching relationship;

[0070] Set the planar distance threshold and the spatial distance threshold. According to the model parameters, obtain the coordinate position of the corner point corresponding to any vehicle-mounted surround-view camera in the same group mapped onto the parking scene image data collected by the front-view fisheye camera, and then obtain the coordinate distance between the corresponding two corner points in the same parking scene image data;

[0071] If the coordinate distance is greater than or equal to the planar distance threshold, do nothing;

[0072] If the coordinate distance is less than the planar distance threshold, mark the corresponding two corner points as inliers;

[0073] Furthermore, according to the inlier correspondence relationship between each parking scene image data, obtain the internal parameter matrix (focal length f x , f y , principal point offset c x , c y ) and the external parameter matrix (rotation matrix R, translation vector T) of each camera through the Zhang Zhengyou calibration method;

[0074] Obtain the world coordinate positions of each inlier through the internal parameter matrix and the external parameter matrix of each camera. The world coordinate positions represent the actual spatial positions of each inlier according to the hardware parameters of the camera;

[0075] Map the world coordinate positions of the inliers corresponding to each vehicle-mounted surround-view camera onto the spatial calibration board, and judge the spatial distance between the coordinate positions of each inlier on the spatial calibration board and the corresponding world coordinate positions. If the spatial distance is greater than or equal to the spatial distance threshold, adjust the hardware parameters of the corresponding vehicle-mounted surround-view camera until the spatial distance is less than the spatial distance threshold.

[0076] Furthermore, after the hardware parameters of each vehicle-mounted surround-view camera adjacent to the front-view fisheye camera are adjusted, use the same method to adjust the hardware parameters of the remaining two vehicle-mounted surround-view cameras according to the parking scene image data collected by each vehicle-mounted surround-view camera adjacent to the front-view fisheye camera;

[0077] After the hardware parameters of all on-vehicle surround-view cameras are adjusted, each camera continuously collects parking scene image data within its shooting range. According to the inlier correspondence between the on-vehicle surround-view cameras and the front fisheye camera, the parking scene image data collected at each same time node is stitched together to obtain a multi-view scene image.

[0078] Further, the step S2 is implemented through the following process:

[0079] Based on the multi-view scene image, a dynamic parking scene model of the parking scene is established, and a three-dimensional coordinate system is established with the vehicle as the origin. Then, the dynamic parking scene model is mapped into the three-dimensional coordinate system.

[0080] The dynamic parking scene model is divided into several model voxels of the same size. The model voxels are cubic, and each face has an independent planar pixel value.

[0081] Starting from the model voxels at the eight vertex positions of the dynamic parking scene model simultaneously, the pixel averaging of each model voxel with its adjacent model voxels is performed, that is, the pixels on the same face of the model voxel and the adjacent model voxels are added and then the average value is taken, and the average value is reassigned to the corresponding face. No operation is performed on the positions where the model voxel has no adjacent face.

[0082] Repeat the above model voxel pixel averaging operation until the model voxels in the eight directions converge to the same position for model voxel pixel averaging.

[0083] Among them, the eight vertex positions are the front, rear, left, right, left front, left rear, right front, and right rear respectively.

[0084] When the dynamic parking scene model completes the model voxel pixel averaging, the Gaussian mixture clustering algorithm is used to cluster the pixels of each voxel. Then, several scene object models are divided in the dynamic parking scene model, and scene semantic labels are set for each scene object model. The scene semantic labels include pedestrians, vehicles, cones, curbs, parking spaces, etc.

[0085] Set the scene model update period. The frequency duration of the scene model update period is generally 10 ms. Randomly calibrate n motion feature points for each scene object model. Then, whenever a scene model update period ends, detect whether the motion feature points have displacement. n is a natural number greater than 0.

[0086] If the displacement of the motion feature points is detected in three consecutive scene model update periods, the corresponding scene object model is recorded as a dynamic object scene model; otherwise, it is recorded as a static scene object model.

[0087] Further, the step S3 is implemented through the following process:

[0088] The user selects a target parking position in the dynamic parking scenario model according to the semantic tags, and then uses the improved rapidly-exploring random tree algorithm to mark the current position of the vehicle as the starting point and the target parking position as the ending point in the dynamic parking scenario model. Each dynamic object scenario model and static scenario object model are regarded as obstacles, and a collision-free parking path is traversed in the dynamic parking scenario model.

[0089] At the start time of each scene model update cycle, the prediction uncertainty is quantified according to Gaussian process regression, and the spatial positions of each dynamic object scenario model in the next scene model update cycle are predicted. The formula is:

[0090] ;

[0091] where represents the predicted spatial coordinates of the motion feature points of the corresponding dynamic object scenario model in the dynamic parking scenario model at the t-th scene model update cycle, , , represent the predicted position means in the x-axis, y-axis, and z-axis directions of the three-dimensional Gaussian distribution, represents the predicted position standard deviation values in each direction;

[0092] The predicted spatial coordinates of the motion feature points of each dynamic object scenario model are mapped to the displacement positions of each dynamic object scenario model in the dynamic parking scenario model, and the displacement directions are marked.

[0093] Furthermore, the step S4 is implemented through the following process:

[0094] Whenever the displacement positions of each dynamic object scenario model in the next scene model update cycle are updated in the dynamic parking scenario model, it is judged whether the vehicle will collide in the next scene model update cycle along the collision-free parking path;

[0095] If it is judged that no collision occurs, no operation is performed;

[0096] If it is judged that a collision occurs, multiple detour collision-free parking paths are regenerated with the current position of the vehicle as the starting point and the target parking position as the ending point, and each detour collision-free parking path is divided into several driving path nodes according to the current driving speed of the vehicle and the time length of the scene model update cycle;

[0097] And the real-time risk values of each detour collision-free parking path are calculated;

[0098] The calculation formula of the real-time risk value is:

[0099] ;

[0100] ;

[0101] ;

[0102] ;

[0103] Among them, represents the real-time risk value of each collision-free parking path around the detour, , , represent weight parameters, represents the risk influencing factors corresponding to the motion feature points, represents the risk influencing factors corresponding to the length of the parking path, represents the risk influencing factors corresponding to the change rate of the steering angle, , , , , , represent the coordinate values of the i-th driving path node and the coordinate values of the nearest motion feature point on the nearest dynamic object scene model. L represents the total length of the corresponding collision-free parking path around the detour, represents the change rate of the steering angle of the i-th driving path node, and N represents the total number of driving path nodes;

[0104] Set a real-time risk threshold, compare the real-time risk value of the collision-free parking path around the detour with the smallest real-time risk value. If the smallest real-time risk value is less than or equal to the real-time risk threshold, then replace the original dynamic collision-free parking path with the corresponding collision-free parking path around the detour;

[0105] If the smallest real-time risk value is greater than the real-time risk threshold, then there is no executable parking path currently;

[0106] Repeat the above determination process every time a scene model update cycle passes until the vehicle reaches the target parking position.

[0107] The above embodiments are only used to illustrate the technical method of the present invention and not to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that the technical method of the present invention can be modified or equivalently replaced without departing from the spirit and scope of the technical method of the present invention.

Claims

1. An automatic parking obstacle avoidance method based on vehicle-mounted image data, characterized in that It includes the following steps: Step S1: Establish a spatial calibration board, map the parking scene image data collected by the vehicle surround-view cameras and the front fisheye camera onto the spatial calibration board, mark several corner points therein, match the corner points of different parking scene image data with each other, record some corner points as inliers according to the matching results, then adjust the hardware parameters of the vehicle surround-view cameras according to the inlier positions, and splice the parking scene image data collected by the adjusted vehicle surround-view cameras and the front fisheye camera to obtain a multi-view scene image; Step S2: Establish a dynamic parking scene model centered on the vehicle according to the multi-view scene image, divide several scene object models in the dynamic parking scene model, and set scene semantic labels for each scene object model; Step S3: Select a target parking position in the dynamic parking scene model, generate a corresponding collision-free parking path and set a scene model update period, and then obtain the spatial positions of the scene object models in the next scene model update period; Step S4: According to the spatial positions of the scene object models in the next scene model update period, judge whether the collision-free parking path will collide, generate multiple detour collision-free parking paths according to the judgment results, obtain the real-time risk values of each detour collision-free parking path, and then select the detour collision-free parking path with the smallest real-time risk value as the new collision-free parking path and execute it.

2. The automatic parking obstacle avoidance method based on vehicle-mounted image data according to claim 1, wherein The process of marking corner points in the parking scene image data includes: Deploy four vehicle surround-view cameras and one front fisheye camera on the vehicle. Each vehicle surround-view camera and the front fisheye camera are distributed according to spatial positions, and there is an overlapping part in the shooting ranges of cameras at adjacent spatial positions; When the vehicle generates a parking request, make each vehicle surround-view camera and one front fisheye camera simultaneously capture parking scene image data from different angles and establish a spatial calibration board; Map each parking scene image data onto the spatial calibration board according to the camera and the shooting angle, and mark several corner points in each parking scene image data through OpenCV.

3. The automatic parking obstacle avoidance method based on vehicle-mounted image data according to claim 2, wherein The process of matching the corner points of different parking scene image data with each other includes: Taking the parking scene image data captured by the front fisheye camera as the reference image data, match the corner points in the parking scene image data captured by the front fisheye camera and the vehicle surround-view cameras at its adjacent positions with each other; If it is judged that two corner points correspond to the same position, it is judged that the two corner points are matched, otherwise it is judged that they are not matched, and the mutually matched corner points are grouped, and each group contains several corner points.

4. The automatic parking obstacle avoidance method based on vehicle-mounted image data according to claim 3, wherein The process of adjusting the hardware parameters of each vehicle surround-view camera includes: Obtain the model parameters of each group of corner points. The model parameters are used to represent the coordinate position parameters of the corner points of different parking scene image data mapped onto another parking scene image data according to the matching relationship. Set a planar distance threshold and a spatial distance threshold, obtain the corner points corresponding to any one of the on-vehicle surround-view cameras in the same group according to the model parameters, map them to the coordinate positions of the parking scene image data collected by the front-view fisheye camera, and then obtain the coordinate distance between the corresponding two corner points in the same parking scene image data; If the coordinate distance is greater than or equal to the planar distance threshold, no operation is performed. If the coordinate distance is less than the planar distance threshold, the corresponding two corner points are marked as inliers, and then the intrinsic matrix and extrinsic matrix of each camera are obtained according to the inlier correspondence relationship between the parking scene image data; Obtain the world coordinate positions of each inlier through the intrinsic matrix and extrinsic matrix of each camera, map the world coordinate positions of the inliers corresponding to each on-vehicle surround-view camera to the spatial calibration board, and judge the spatial distance between the coordinate positions of each inlier on the spatial calibration board and the corresponding world coordinate positions. If the spatial distance is greater than or equal to the spatial distance threshold, adjust the hardware parameters of the corresponding on-vehicle surround-view camera until the spatial distance is less than the spatial distance threshold; 5. The automatic parking obstacle avoidance method based on vehicle-mounted image data according to claim 4, characterized in that, The process of obtaining the multi-view scene image includes: When the hardware parameters of all on-vehicle surround-view cameras are adjusted, each camera continuously collects the parking scene image data within its shooting range, and splices the parking scene image data collected at each same time node according to the inlier correspondence relationship between the on-vehicle surround-view camera and the front-view fisheye camera, and then obtains the multi-view scene image; 6. The automatic parking obstacle avoidance method based on vehicle-mounted image data according to claim 5, wherein The process of dividing a number of scene object models in the dynamic parking scene model includes: Establish a dynamic parking scene model of the parking scene according to the multi-view scene image, and establish a three-dimensional coordinate system with the vehicle as the origin, and then map the dynamic parking scene model to the three-dimensional coordinate system; Divide the dynamic parking scene model into a number of model voxels of the same size. Starting from the model voxels at the eight vertex positions of the dynamic parking scene model simultaneously, perform pixel averaging on each model voxel and its adjacent model voxels; When the pixel averaging of the model voxels of the dynamic parking scene model is completed, cluster each voxel pixel through the Gaussian mixture clustering algorithm, and then divide a number of scene object models in the dynamic parking scene model, and set scene semantic labels for each scene object model; Set a scene model update period, randomly calibrate n motion feature points for each scene object model, and then whenever a scene model update period ends, detect whether the motion feature points have displacement, where n is a natural number greater than 0; If the displacement of the motion feature points is detected in three consecutive scene model update periods, mark the corresponding scene object model as a dynamic object scene model, otherwise mark it as a static scene object model; 7. The automatic parking obstacle avoidance method based on vehicle-mounted image data according to claim 6, wherein, The process of obtaining the spatial position of the scene object model in the next scene model update period includes: Select a target parking position in the dynamic parking scene model, mark the current position of the vehicle as the starting point and the target parking position as the ending point in the dynamic parking scene model, and each scene object model as an obstacle, and traverse a collision-free parking path in the dynamic parking scene model; At the start of each scene model update cycle, predict the spatial positions of the scene models of dynamic objects in the next scene model update cycle; Map the predicted spatial coordinates of the motion feature points of each dynamic object scene model to the displacement positions of each dynamic object scene model in the dynamic parking scene model.

8. The automatic parking obstacle avoidance method based on vehicle-mounted image data according to claim 7, characterized in that, The generation process of the detour collision-free parking path includes: Whenever the dynamic parking scene model updates the displacement positions of each dynamic object scene model in the next scene model update cycle, determine whether the vehicle will collide in the next scene model update cycle along the collision-free parking path; Regenerate multiple detour collision-free parking paths according to the judgment result, and divide each detour collision-free parking path into several driving path nodes based on the current driving speed of the vehicle and the time length of the scene model update cycle, and obtain the real-time risk values of each detour collision-free parking path.

9. The automatic parking obstacle avoidance method based on vehicle-mounted image data according to claim 8, wherein The process of obtaining the real-time risk values of each detour collision-free parking path includes: Set a real-time risk threshold, compare the real-time risk values of the detour collision-free parking paths with the smallest real-time risk value. If the smallest real-time risk value is less than or equal to the real-time risk threshold, replace the original dynamic collision-free parking path with the corresponding detour collision-free parking path; If the smallest real-time risk value is greater than the real-time risk threshold, there is no executable parking path at present.

Images