Space detection method and system for three-dimensional parking space and storage medium

Abstract

The invention provides a space detection method and system for a three-dimensional parking stall and a storage medium, and the method comprises the steps: firstly obtaining the multi-frame image information of the three-dimensional parking stall, and carrying out the detection of the front slope point of the parking stall in addition to the vehicle exterior point in the space detection process of the three-dimensional parking stall; the method comprises the following steps: firstly, acquiring a plurality of frames of image information, then detecting parking space points and front slope points of each frame of image information of a model through angular points, carrying out inter-frame association on each frame of parking space points and front slope points to obtain a plurality of pairs of optimal matching point pairs, and finally, based on the plurality of pairs of optimal matching point pairs and inertial navigation values, calculating to obtain final space three-dimensional coordinates of the three-dimensional parking space and the parking space front slope. Through the corner detection model, the stability and accuracy of key point detection are improved. And through joint optimization of multi-frame results, more accurate spatial three-dimensional coordinates of the three-dimensional parking space and the parking space front slope can be obtained.

Description

Technical Field

[0001] The present application relates to the field of assisted driving technology, and in particular to a spatial detection method, system and storage medium for a three-dimensional parking space. Background Art

[0002] With the acceleration of urbanization, the problem of parking difficulties has become increasingly prominent. In order to solve this problem, intelligent parking systems have emerged. Among them, three-dimensional parking spaces are widely used in various parking places due to their high space utilization and large parking capacity. However, the 3D spatial position detection of three-dimensional parking spaces is the key to realizing automatic parking. Traditional detection methods mainly rely on ultrasonic sensors, infrared sensors, etc., but these methods are easily affected by environmental factors and have low detection accuracy.

[0003] In order to improve the detection accuracy of three-dimensional parking spaces, detection methods based on computer vision and deep learning have gradually been applied in recent years. These methods mainly collect parking space images through cameras and then use deep learning algorithms to perform image-based target detection or segmentation.

[0004] For example, some methods use convolutional neural networks to detect targets in parking space images and determine whether the parking space is occupied. However, rectangular frames are generally used, and the position detection accuracy of irregular targets is poor. Some other methods use semantic segmentation technology to separate the parking space area from the background to achieve accurate positioning of the parking space. However, this method classifies each pixel, which requires a lot of calculations, and the accuracy of fine parts such as corner points and contours of the parking space is not high. At the same time, subsequent complex post-processing algorithms are still required to calculate the parking space position.

[0005] Moreover, the detection accuracy of this pure image detection method depends on the flatness of the ground and the parking space being at the same height as the ground. However, three-dimensional parking spaces often have a slope, and the parking plane of three-dimensional parking spaces is at a certain height from the ground, resulting in low accuracy of parking space corner points in pure target detection, and it is impossible to meet the high-precision detection and positioning requirements of narrow three-dimensional parking spaces. Summary of the invention

[0006] In order to solve the above technical problems, the present application provides a spatial detection method, system and storage medium for a three-dimensional parking space.

[0007] In a first aspect, the present application provides a spatial detection method for a three-dimensional parking space, the method specifically comprising the following steps: S1: The vehicle-mounted camera selects a three-dimensional parking space and continuously shoots it to obtain multiple frames of image information of the three-dimensional parking space.

[0008] S2: Perform distortion correction processing on each frame of image information to obtain processed three-dimensional parking space images of each frame.

[0009] S3: Based on the corner point detection model, key points in each frame of the three-dimensional parking space image are extracted to obtain the parking position point and the parking space front slope point in each frame.

[0010] S4: perform frame-to-frame association on the parking spots and the front slope points of the parking spots in each frame to obtain multiple pairs of best matching points.

[0011] S5: Based on the multiple pairs of best matching points and the inertial navigation values, the final spatial three-dimensional coordinates of the three-dimensional parking space and the front slope of the parking space are calculated.

[0012] The spatial detection method for a three-dimensional parking space proposed in the present application, in the process of processing the three-dimensional parking space, not only detects the parking position point, but also detects the front slope point of the parking space. Compared with the traditional rectangular frame detection, the spatial detection method proposed in the present application is more flexible and the actual imaging effect is more realistic.

[0013] Furthermore, the multiple frames of image information include three-dimensional parking spaces at different positions and angles.

[0014] Through this method design, multiple frames of image information of three-dimensional parking spaces at different positions and angles can be collected to prepare for subsequent multi-frame optimization.

[0015] Furthermore, the construction process of the corner detection model is as follows: S21: Collect scene image data including the three-dimensional parking space and the slope in front of the parking space.

[0016] S22: Perform distortion correction on the scene image data to obtain distortion-corrected image information.

[0017] S23: Train and optimize the deep neural network model through the distortion-corrected image information to obtain a final corner detection model.

[0018] By constructing a corner point detection model, it can automatically identify and locate the four parking spots and four front slope points of the three-dimensional parking space in the two-dimensional image, thereby realizing accurate detection of the three-dimensional parking space in a complex environment.

[0019] Furthermore, the scene image data also includes: feature annotation of image information of the three-dimensional parking space and the front slope of the parking space under different lighting conditions, different angles and different scenes; wherein the feature annotation is to annotate the position, size and direction information of the three-dimensional parking space and the front slope of the parking space.

[0020] Through this method design, the corner detection model can effectively overcome the influence of environmental factors such as bad weather and poor lighting conditions, and improve the stability and accuracy of detection.

[0021] Furthermore, the step S4 is specifically as follows: inter-frame association is performed on the parking spots and front slope points of any two frames until the corresponding associations between the parking spots and front slope points of each frame are completed, thereby obtaining multiple optimal matching point pairs.

[0022] Through this method design, a more reliable matching point pair can be obtained, which is beneficial to improving the detection accuracy in the subsequent three-dimensional parking space detection process.

[0023] Furthermore, the step S5 is specifically as follows: S51: Obtain the vehicle body inertial navigation value, and calculate the rotation matrix and translation vector between the camera positions corresponding to each pair of best matching points.

[0024] S52: Based on the rotation matrix and the translation vector, triangulate each pair of best matching points to obtain the spatial coordinates of the three-dimensional parking space and the front slope of the parking space corresponding to each pair of best matching points.

[0025] S53: jointly optimizing the spatial three-dimensional coordinates of the three-dimensional parking space and the front slope of the parking space corresponding to each pair of best matching points to obtain the final spatial three-dimensional coordinates of the three-dimensional parking space and the front slope of the parking space.

[0026] Through this method design, the spatial coordinates of the three-dimensional parking space and the front slope of the parking space corresponding to each pair of best matching points are obtained, and joint optimization is performed to obtain more accurate spatial coordinates of the three-dimensional parking space and the front slope of the parking space.

[0027] Furthermore, the constraints of the triangulation calculation are: the four corner points of the three-dimensional parking space are at the same height, the two corner points of the front slope of the parking space are grounded, and the other two corner points are at the same height as the three-dimensional parking space.

[0028] Through this design method, according to the prior conditions that the heights of the four parking spaces of the three-dimensional parking space are consistent, the two front slope points of the three-dimensional parking space are grounded, and the other two front slope points of the three-dimensional parking space are at the same height as the three-dimensional parking space, the triangulation calculation is constrained to obtain more accurate final spatial three-dimensional coordinate points of the three-dimensional parking space and the front slope of the parking space.

[0029] Furthermore, after step S5, the method further includes: S54: taking the final spatial coordinates of the three-dimensional parking space and the front slope of the parking space as parameters to participate in the calculation of the automatic parking algorithm to obtain an automatic parking control vector.

[0030] S55: Controlling the vehicle to perform automatic parking using the automatic parking control vector.

[0031] Through this method design, after the final spatial three-dimensional coordinates of the three-dimensional parking space and the front slope of the parking space are obtained, the obtained final spatial three-dimensional coordinates of the three-dimensional parking space and the front slope of the parking space are used in the regulatory parking system of the autonomous driving vehicle to guide the vehicle to perform accurate parking operations in real time.

[0032] In a second aspect, the present application provides a spatial detection system for a three-dimensional parking space, the system specifically comprising: Image information acquisition module: used for the vehicle-mounted camera to select a three-dimensional parking space and continuously shoot it to obtain multiple frames of image information of the three-dimensional parking space.

[0033] Distortion correction processing module: used to perform distortion correction processing on each frame of image information to obtain each frame of processed three-dimensional parking space image.

[0034] Detection module: used to extract key points in each frame of the three-dimensional parking space image based on the corner point detection model, and obtain the parking position point and front slope point of each frame.

[0035] Inter-frame association module: used to perform inter-frame association on the parking spots and front slope points of each frame to obtain multiple pairs of best matching points.

[0036] Calculation module: used for calculating the final spatial three-dimensional coordinates of the three-dimensional parking space and the front slope of the parking space based on the multiple pairs of best matching points and the inertial navigation values.

[0037] Furthermore, after the image information acquisition module obtains multiple frames of image information of the three-dimensional parking space, it sends it to the distortion correction processing module; after the distortion correction processing module obtains the processed three-dimensional parking space image frames, it sends it to the detection module; after the detection module extracts the four corner points of the three-dimensional parking space and the four corner points of the front slope of the parking space in each frame of the three-dimensional parking space image, it sends it to the inter-frame association module; the inter-frame association module obtains multiple pairs of one-to-one corresponding best matching point pairs, and sends them to the calculation module; after the calculation module calculates the final spatial three-dimensional coordinates of the three-dimensional parking space and the front slope of the parking space, it sends it to the control system of the vehicle.

[0038] On the third aspect, based on the same inventive concept, the present application also proposes a computer-readable storage medium, which stores computer-executable instructions. When the computer-executable instructions are executed by a control processor, the spatial detection method of the three-dimensional parking space as described above is implemented.

[0039] It should be noted that since the computer-readable storage medium in this embodiment and the battery cluster management method in the above embodiment are based on the same inventive concept, the corresponding contents in the method embodiment are also applicable to the system embodiment and will not be described in detail here.

[0040] In summary, the present application proposes a spatial detection method, system and storage medium for a three-dimensional parking space. The spatial detection method first obtains multiple frames of image information of the three-dimensional parking space, then detects the parking position and front slope points of each frame of the image information of the model through corner point detection, and performs frame-to-frame correlation on the parking position and front slope points of each frame to obtain multiple pairs of best matching point pairs, and finally calculates the final spatial three-dimensional coordinates of the three-dimensional parking space and the front slope of the parking space based on the multiple pairs of best matching point pairs and inertial navigation values. The present application also proposes a spatial detection system and a storage medium for implementing the spatial detection method.

[0041] Compared with the prior art, this application has at least the following beneficial effects: The spatial detection method for three-dimensional parking spaces proposed in this application, in the process of processing three-dimensional parking spaces, not only detects the parking spot, but also detects the front slope point of the parking space. Compared with the traditional rectangular frame detection, the spatial detection method proposed in this application is more flexible and the actual imaging effect is more realistic. The spatial detection method for three-dimensional parking spaces proposed in this application can enhance the detection capability of ramps and improve the accuracy of detection. This method can not only reduce the detection range limitations and environmental factors of solutions based on ultrasonic and infrared sensors, but also avoid the high equipment cost of solutions based on lidar, which is conducive to large-scale application. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] Figure 1 It is a schematic flow chart of a spatial detection method for a three-dimensional parking space shown in an embodiment of the present application.

[0043] Figure 2 It is a schematic diagram of the structure of a space detection system for a three-dimensional parking space shown in an embodiment of the present application. DETAILED DESCRIPTION

[0044] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0045] Embodiment 1: As attached Figure 1 As shown, the present application provides a spatial detection method for a three-dimensional parking space, and the specific steps of the method include: S1: The vehicle-mounted camera selects a three-dimensional parking space and continuously shoots it to obtain multiple frames of image information of the three-dimensional parking space.

[0046] S2: Perform distortion correction processing on each frame of image information to obtain processed three-dimensional parking space images of each frame.

[0047] S3: Based on the corner point detection model, key points in each frame of the three-dimensional parking space image are extracted to obtain the parking position point and the parking space front slope point in each frame.

[0048] S4: perform frame-to-frame association on the parking spots and the front slope points of the parking spots in each frame to obtain multiple pairs of best matching points.

[0049] S5: Based on the multiple pairs of best matching points and the inertial navigation values, the final spatial three-dimensional coordinates of the three-dimensional parking space and the front slope of the parking space are calculated.

[0050] In the embodiment of the present invention, optionally, the spatial detection method of the three-dimensional parking space can be applied to vehicles such as sedans, SUVs or any vehicles equipped with a camera and an intelligent assisted driving system.

[0051] In an embodiment of the present invention, optionally, the vehicle obtains multiple frames of basic image information of the three-dimensional parking space through a vehicle-mounted camera, and then the internal function of the vehicle auxiliary driving system obtains the final spatial three-dimensional coordinates of the three-dimensional parking space and the front slope of the parking space, and controls the vehicle to complete automatic parking.

[0052] In the embodiment of the present invention, optionally, the multiple frames of image information include three-dimensional parking spaces at different positions and different angles.

[0053] In the embodiment of the present invention, optionally, the positions and rotation angles of the vehicle-mounted camera corresponding to any two frames of image information are different. During the process of continuously photographing the three-dimensional parking space, the vehicle-mounted camera can be displaced forward and backward or left and right, and can also be rotated counterclockwise or clockwise, or multiple vehicle-mounted cameras can be used to photograph the three-dimensional parking space from multiple angles, but not limited to this. It is only necessary to ensure that the camera positions corresponding to any two frames of image information of the three-dimensional parking space are different.

[0054] In an embodiment of the present invention, optionally, the construction process of the corner detection model is specifically: collecting a large amount of distortion-corrected image information including three-dimensional parking spaces and front slopes of parking spaces, and training a deep neural network based on the distortion-corrected image information to obtain a corner detection model.

[0055] In the embodiment of the present invention, optionally, the corner detection model may be a deep neural network (DNN), which is a special type of neural network with multiple hidden layers. Some common deep neural network architectures include: Multilayer Perceptron (MLP): The simplest deep neural network, consisting of multiple fully connected hidden layers.

[0056] Convolutional Neural Network (CNN): Mainly used for image recognition and computer vision tasks, including convolutional layers, pooling layers and fully connected layers.

[0057] In an embodiment of the present invention, optionally, the deep neural network (DNN) uses a convolutional neural network.

[0058] In the embodiment of the present invention, optionally, the large amount of distortion-corrected image information including the three-dimensional parking space and the slope in front of the parking space is specifically: distortion-corrected image information including the three-dimensional parking space and the slope in front of the parking space under different lighting conditions, different angles and different scenes.

[0059] In the embodiment of the present invention, optionally, the large amount of distortion-corrected image information including the three-dimensional parking space and the slope in front of the parking space needs to cover different lighting conditions, angles and scenes. The distortion-corrected image information is used to train a deep neural network, and after multiple rounds of iterations and fine-tuning, a deep neural network model that can automatically identify and output the corner point positions of the three-dimensional parking space and the slope in front of the parking space, that is, a corner point detection model, is obtained.

[0060] In the embodiment of the present invention, optionally, the step S4 is specifically as follows: each frame parking spot and front slope point is provided with corresponding unique semantic information, and based on the semantic information, any two frames of parking spots and front slope points are inter-frame associated, until the corresponding associations between the parking spots and front slope points of each frame are completed, and multiple best matching point pairs are obtained.

[0061] In the embodiment of the present invention, optionally, the semantic information refers to that in the process of constructing the corner point detection model, the three-dimensional parking space and the front slope of the parking space are combined into a group of target detections, and each group has a unique group number.

[0062] In the embodiment of the present invention, optionally, based on the group number, the four corner points of the three-dimensional parking space and the four corner points of the front slope of the parking space are further divided into two types of information, and each is sorted clockwise. That is, each of the eight corner points has its own type information (belonging to the three-dimensional parking space corner point or the front slope corner point of the parking space) and a unique number (one of 1, 2, 3, and 4 after clockwise sorting), and the inter-frame association between the two frames of image information corresponds one to one according to the type and number of the corner point.

[0063] In the embodiment of the present invention, optionally, the step S5 is specifically: S51: Obtain the vehicle body inertial navigation value, and calculate the rotation matrix and translation vector between the camera positions corresponding to each pair of best matching points.

[0064] S52: Based on the rotation matrix and the translation vector, triangulate each pair of best matching points to obtain the spatial coordinates of the three-dimensional parking space and the front slope of the parking space corresponding to each pair of best matching points.

[0065] S53: jointly optimizing the spatial three-dimensional coordinates of the three-dimensional parking space and the front slope of the parking space corresponding to each pair of best matching points to obtain the final spatial three-dimensional coordinates of the three-dimensional parking space and the front slope of the parking space.

[0066] In the embodiment of the present invention, the displacement and angle of the vehicle-mounted camera between each pair of best matching points can be obtained by inertial navigation values, the angle is converted into a rotation matrix R, the size is [3*3], the translation is converted into a translation vector T, the size is [1*3], and the translation vector T and the rotation matrix R are combined into a transformation matrix of 4*4, which is the posture of the vehicle-mounted camera. In this way, the transformation matrix obtained by using the sensor is more accurate than the transformation matrix obtained by using the feature points, and it is not easy to diverge.

[0067] In the embodiment of the present invention, triangulation calculation refers to the process of determining the position or distance of an unknown point by using the principle of triangulation. It is widely used in the fields of geographic information system (GIS), computer vision, surveying, wireless positioning, etc. The main methods include: Triangle Similarity Method (Triangulation Method): This is the most basic triangulation method. By placing at least three measuring instruments (or observation points) at known locations, and then measuring the angles or distances between the unknown target point and these known points, the position of the unknown point is calculated using the geometric relationship of the triangle.

[0068] Triangulation by Range Finding: This method uses the distance between the target point and at least three known locations to determine the location of the target point through mathematical calculations. This method is common in the fields of radar, laser ranging, GPS, etc.

[0069] Triangulation by Angulation: Determine the position of a target point by measuring the angles between the target point and at least three known positions. This method is common in wireless positioning and electronic navigation.

[0070] In practical applications, triangulation involves precise measurement techniques and mathematical calculation methods, such as trigonometric functions, vector calculations, numerical optimization, etc. The accuracy and precision of the results depend on the precision of the measuring instruments, the quality of the observation conditions, and the effectiveness of the calculation methods.

[0071] In an embodiment of the present invention, optionally, the parking spots and front slope points detected by the corner point detection model will be regarded as feature points, and each detected feature point will be described as a vector or feature descriptor to describe the image information around it. The assisted driving system will try to match the different detected feature points, which usually involves calculating the similarity or distance between the feature points to determine whether they are feature points from the same object. Once the feature points are successfully matched, the assisted driving system can use their projected positions under different camera perspectives to perform triangulation calculations.

[0072] In the embodiment of the present invention, optionally, for each pair of matched feature points, the assisted driving system knows the pixel coordinates (i.e., projection positions) of the feature points in different camera images, as well as the camera's intrinsic parameters (such as focal length, principal point) and extrinsic parameters (the position and orientation of the camera in space). Based on the parallax (the difference in viewing angle between different camera images) and the camera parameters, the assisted driving system can calculate the three-dimensional position of the object represented by the feature point in space.

[0073] In the embodiment of the present invention, optionally, the calculation of the three-dimensional position of the feature point in space is achieved by parallax and camera parameters, and the specific steps are as follows: Parallax calculation: Disparity refers to the difference in the position of feature points of the same object under different camera views. It can be determined by calculating the distance between the pixel coordinates of the feature points in different cameras.

[0074] Suppose there are two cameras, each capturing the same feature point. The pixel coordinates of the feature point in the first camera are ( (u_1, v_1) ), and the pixel coordinates of the feature point in the second camera are ( (u_2, v_2) ).

[0075] The parallax (d) can be calculated using the following formula: [d = u_1 - u_2] Here we assume that the cameras are horizontally aligned, so the parallax is mainly determined by the difference in the (u) coordinates.

[0076] Triangulation calculation: Once the disparity ( d ) is obtained, the following parameters can be used for triangulation calculation: Internal parameters of the camera (such as focal length, principal point position).

[0077] The camera's extrinsic parameters (such as the camera's position and orientation in space).

[0078] Intrinsic parameters are usually obtained through a camera calibration process and they describe the perspective projection characteristics of the camera.

[0079] The external parameters describe the position and orientation of the camera in space, which are usually determined by the accurate position and orientation information when the camera is installed.

[0080] Triangulation process: When the disparity ( d ) and camera parameters are known, the position of feature points in space can be calculated using the principle of triangulation.

[0081] The basic principle of triangulation utilizes the concept of similar triangles to determine the coordinates of an object in space based on pixel coordinates, internal and external parameters, and parallax.

[0082] Generally speaking, the 3D position of an object is obtained by converting pixel coordinates into normalized coordinates in the camera coordinate system and then applying an inverse projection matrix to map the normalized coordinates to the world coordinate system.

[0083] In an embodiment of the present invention, optionally, multi-frame joint optimization is the process of solving the position of a 3D point in the world coordinate system under the condition of known 2D observations of multiple frames of images, the camera intrinsic parameter matrix, and the camera pose of each observed image frame. In an embodiment of the present invention, optionally, the number of frames is at least greater than or equal to 2.

[0084] In the embodiment of the present invention, optionally, the constraint conditions of the triangulation calculation are: the four corner points of the three-dimensional parking space are at the same height, the two corner points of the front slope of the parking space are grounded, and the other two corner points are at the same height as the three-dimensional parking space.

[0085] In the embodiment of the present invention, optionally, based on the prior conditions that the heights of the four parking spots of the three-dimensional parking space are consistent, the two front slope points of the three-dimensional parking space are grounded, and the other two front slope points of the three-dimensional parking space are at the same height as the three-dimensional parking space, the triangulation calculation is constrained to obtain more accurate final spatial three-dimensional coordinate points of the three-dimensional parking space and the front slope of the parking space.

[0086] In the embodiment of the present invention, optionally, after step S5, the following further comprises: The final spatial three-dimensional coordinates of the three-dimensional parking space and the front slope of the parking space are used as parameters to participate in the calculation of the automatic parking algorithm to obtain an automatic parking control vector.

[0087] The vehicle is controlled to perform automatic parking using the automatic parking control vector.

[0088] In an embodiment of the present invention, optionally, ROS (Robot Operating System) is used as middleware to send the estimated final spatial coordinates of the three-dimensional parking space and the front slope of the parking space to the control system of the vehicle. The control system adjusts the driving path and speed of the vehicle in real time according to this information to ensure that the vehicle can be parked safely and accurately in the three-dimensional parking space, but not limited to this. The middleware can also be other devices or systems that can store and send data.

[0089] In the embodiment of the present invention, optionally, the obtaining of the automatic parking control vector is specifically: According to the spatial coordinates of the current vehicle position, the final spatial coordinates of the three-dimensional parking space and the front slope of the parking space, a path planning algorithm is used to generate a safe path from the current position to the target parking space, wherein the path planning algorithm may include: A* Algorithm: The A* algorithm is a common heuristic search algorithm used to find the shortest path from a start point to an end point in a graph or grid. It combines breadth-first search with heuristic evaluation (such as Manhattan distance or Euclidean distance) to speed up the search process while ensuring that the optimal path is found.

[0090] Dijkstra Algorithm: Dijkstra's algorithm is a classic shortest path algorithm in unweighted graphs. It gradually determines the shortest path by gradually expanding the nodes closest to the starting point. In path planning, if the weight of the road section (such as distance or other costs) is taken into account, a weighted version of Dijkstra's algorithm can be used.

[0091] RRT algorithm (Rapidly-exploring Random Tree): The RRT algorithm is often used for path planning in non-precise environments, such as rapid exploration and path generation in complex dynamic environments. It generates paths through random samples and tree structures, and is suitable for motion planning problems in high-dimensional spaces.

[0092] Depth First Search (DFS) and Breadth First Search (BFS): DFS and BFS are among the simplest path search algorithms and are suitable for simple graph structures or grids. DFS explores along the depth of the path, while BFS expands the search layer by layer. Although they are simple, they may not be as efficient as other algorithms in some cases.

[0093] Dynamic Programming (DP): Dynamic programming is often used to solve optimization problems, including path planning. It breaks down the problem into sub-problems and uses the results of previous calculations to speed up the solution process. In path planning, dynamic programming can be applied to some specific scenarios, such as the shortest path problem in a grid map.

[0094] Genetic Algorithm: Genetic algorithm is an evolutionary computing technique, which is often used for complex path optimization problems. It simulates the process of natural selection and searches for potential solutions through the evolution of a population. In path planning, genetic algorithms can be used to generate efficient path sets and gradually optimize path quality.

[0095] In practical applications, the selection of an appropriate path planning algorithm depends on the specific requirements of the problem, including environmental complexity, path constraints, and vehicle performance requirements. Modern automatic parking systems usually combine multiple algorithms and strategies to cope with various complex driving scenarios and real-time changing environmental conditions.

[0096] Through this method design, after the final spatial three-dimensional coordinates of the three-dimensional parking space and the front slope of the parking space are obtained, the obtained final spatial three-dimensional coordinates of the three-dimensional parking space and the front slope of the parking space are used in the regulatory parking system of the autonomous driving vehicle to guide the vehicle to perform accurate parking operations in real time.

[0097] Embodiment 2: As attached Figure 2 As shown, the present application provides a spatial detection system for a three-dimensional parking space, and the system specifically includes: Image information acquisition module: used for the vehicle-mounted camera to select a three-dimensional parking space and continuously shoot it to obtain multiple frames of image information of the three-dimensional parking space.

[0098] Distortion correction processing module: used to perform distortion correction processing on each frame of image information to obtain each frame of processed three-dimensional parking space image.

[0099] Detection module: used to extract key points in each frame of the three-dimensional parking space image based on the corner point detection model, and obtain the parking position point and front slope point of each frame.

[0100] Inter-frame association module: used to perform inter-frame association on the parking spots and front slope points of each frame to obtain multiple pairs of best matching points.

[0101] Calculation module: used for calculating the final spatial three-dimensional coordinates of the three-dimensional parking space and the front slope of the parking space based on the multiple pairs of best matching points and the inertial navigation values.

[0102] In an embodiment of the present invention, optionally, the image information acquisition module includes a vehicle-mounted camera. After the vehicle-mounted camera captures the image information, the image information is sent to the distortion correction processing module via the vehicle-mounted CAN or other data transmission methods; the distortion correction processing module includes a processor, which can be a computer, a single-chip microcomputer or an FPGA, and the processor has a built-in image processing algorithm based on the internal calibration parameter setting of the vehicle-mounted camera. After processing each frame of image information, the processed frames of the three-dimensional parking space image are sent to the detection module via the vehicle-mounted CAN or other data transmission methods; the detection module includes a processor, which can be a computer, a single-chip microcomputer or an FPGA, and the processor has a built-in corner detection module. Type; the inter-frame association module includes a processor, which can be a computer, a single-chip microcomputer or an FPGA. The processor has a built-in algorithm for inter-frame association of any two frames of three-dimensional parking space images to obtain multiple pairs of one-to-one corresponding best matching point pairs. The inter-frame association module sends the multiple pairs of one-to-one corresponding best matching point pairs to the calculation module through the vehicle-mounted CAN or other data transmission methods; the calculation module uses the inertial navigation value obtained by the vehicle-mounted CAN and combines the best matching point pairs to obtain the final spatial three-dimensional coordinates of the three-dimensional parking space and the front slope of the parking space. The calculation module includes a processor, which can be a computer, a single-chip microcomputer or an FPGA. The processor has a built-in rotation matrix algorithm, a translation vector algorithm and a triangulation algorithm.

[0103] Embodiment three: Based on the same inventive concept, the present application also proposes a computer-readable storage medium, wherein the computer-readable storage medium stores computer-executable instructions, and when the computer-executable instructions are executed by a control processor, the spatial detection method of the three-dimensional parking space as described above is implemented.

[0104] It should be noted that since the computer-readable storage medium in this embodiment and the battery cluster management method in the above embodiment are based on the same inventive concept, the corresponding contents in the method embodiment are also applicable to the system embodiment and will not be described in detail here.

[0105] In summary, the present application proposes a spatial detection method, system and storage medium for a three-dimensional parking space. The spatial detection method first obtains multiple frames of image information of the three-dimensional parking space, then detects the parking position and front slope points of each frame of the image information of the model through corner point detection, and performs frame-to-frame correlation on the parking position and front slope points of each frame to obtain multiple pairs of best matching point pairs, and finally calculates the final spatial three-dimensional coordinates of the three-dimensional parking space and the front slope of the parking space based on the multiple pairs of best matching point pairs and inertial navigation values. The present application also proposes a spatial detection system and a storage medium for implementing the spatial detection method.

[0106] The spatial detection method for three-dimensional parking spaces proposed in this application, in the process of processing three-dimensional parking spaces, not only detects the parking spot, but also detects the front slope point of the parking space. Compared with the traditional rectangular frame detection, the spatial detection method proposed in this application is more flexible and the actual imaging effect is more realistic. The spatial detection method for three-dimensional parking spaces proposed in this application can enhance the detection capability of ramps and improve the accuracy of detection. This method can not only reduce the detection range limitations and environmental factors of solutions based on ultrasonic and infrared sensors, but also avoid the high equipment cost of solutions based on lidar, which is conducive to large-scale application.

[0107] In several embodiments provided in the present application, it is understood that each box in the flow chart or block diagram can represent a module, a program segment or a part of a code, and the module, a program segment or a part of a code contains one or more executable instructions for realizing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the box can also occur in a different order from that marked in the accompanying drawings. For example, two consecutive boxes can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved.

[0108] If the function is implemented in the form of a software function module and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application can essentially or in other words, the part that contributes to the prior art or the part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for an electronic device to execute all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk and other media that can store program code.

[0109] The specific embodiments described above further describe the purpose, technical solutions and beneficial effects of the present application in detail. It should be understood that the above description is only a specific embodiment of the present application and is not intended to limit the scope of protection of the present application. It is particularly pointed out that for those skilled in the art, any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application should be included in the scope of protection of the present application.

Claims

1. A spatial detection method for a three-dimensional parking space, characterized in that: The method specifically comprises the following steps: S1: Select a three-dimensional parking space and continuously shoot it to obtain multiple frames of image information of the three-dimensional parking space; S2: performing distortion correction processing on each frame of image information to obtain processed three-dimensional parking space images of each frame; S3: extracting key points in each frame of the three-dimensional parking space image based on a corner point detection model to obtain parking spots and front slope points in each frame; S4: perform frame-to-frame association on the parking spots and front slope points of each frame to obtain multiple pairs of best matching points; S5: Based on the multiple pairs of best matching points and the inertial navigation values, the final spatial three-dimensional coordinates of the three-dimensional parking space and the front slope of the parking space are calculated.

2. A spatial detection method for a three-dimensional parking space according to claim 1, characterized in that: The corner point detection model is specifically: S21: collecting scene image data including a three-dimensional parking space and a slope in front of the parking space; S22: performing distortion correction on the scene image data to obtain distortion-corrected image information; S23: Train and optimize the deep neural network model through the distortion-corrected image information to obtain a final corner detection model.

3. The spatial detection method of a three-dimensional parking space according to claim 2, characterized in that: The step S22 further includes: performing feature annotation on the scene image information; wherein the feature annotation is to annotate the position, size and direction information of the three-dimensional parking space and the front slope of the parking space.

4. The spatial detection method of a three-dimensional parking space according to claim 1, characterized in that: The step S4 is specifically: performing inter-frame association between the parking spots and the front slope points of any two frames until the parking spots and the front slope points of each frame are associated with each other in pairs, thereby obtaining a plurality of pairs of best matching points.

5. The spatial detection method of a three-dimensional parking space according to claim 1, characterized in that: The step S5 is specifically: S51: Obtain the vehicle body inertial navigation value, and calculate the rotation matrix and translation vector between the camera positions corresponding to each pair of best matching points; S52: Based on the rotation matrix and translation vector, triangulate each pair of best matching points to obtain the spatial coordinates of the three-dimensional parking space and the front slope of the parking space corresponding to each pair of best matching points; S53: jointly optimizing the spatial three-dimensional coordinates of the three-dimensional parking space and the front slope of the parking space corresponding to each pair of best matching points to obtain the final spatial three-dimensional coordinates of the three-dimensional parking space and the front slope of the parking space.

6. A spatial detection method for a three-dimensional parking space according to claim 5, characterized in that: The constraints of the triangulation calculation are: the heights of the four corner points of the three-dimensional parking space are consistent, the two corner points of the front slope of the parking space are grounded, and the other two corner points are at the same height as the three-dimensional parking space.

7. The spatial detection method of a three-dimensional parking space according to claim 5, characterized in that: After step S5, the method further includes: S54: taking the final spatial coordinates of the three-dimensional parking space and the front slope of the parking space as parameters to participate in the calculation of the automatic parking algorithm, and obtaining an automatic parking control vector; S55: Controlling the vehicle to perform automatic parking using the automatic parking control vector.

8. The spatial detection method of a three-dimensional parking space according to claim 1, characterized in that The multiple frames of image information include three-dimensional parking spaces at different positions and angles.

9. A system for the spatial detection method of a three-dimensional parking space as claimed in claims 1 to 8, characterized in that: Specifically include: Image information acquisition module: used for the vehicle-mounted camera to select a three-dimensional parking space and continuously shoot it to obtain multiple frames of image information of the three-dimensional parking space; Distortion correction processing module: used to perform distortion correction processing on each frame of image information to obtain each frame of processed three-dimensional parking space image; Detection module: based on the corner point detection model, extract the key points in each frame of the three-dimensional parking space image to obtain the parking position and front slope point of each frame; Inter-frame association module: used to perform inter-frame association between parking spots and front slope points in each frame to obtain multiple pairs of best matching points; Calculation module: based on the multiple pairs of best matching points and inertial navigation values, calculate the final spatial three-dimensional coordinates of the three-dimensional parking space and the front slope of the parking space.

10. A computer-readable storage medium storing computer-executable instructions, characterized in that: When the computer executable instructions are executed by the control processor, the spatial detection method of the three-dimensional parking space as described in any one of claims 1-8 is implemented.

Images