Parking lot navigation road network generation method for unmanned AVP scene

Abstract

The invention belongs to the technical field of unmanned driving, and particularly relates to a parking lot navigation road network generation method for an unmanned AVP scene, which comprises the following specific steps: sensor deployment, vehicle data acquisition, map data collection, point cloud data processing, image data processing, data fusion, node determination, edge connection, path planning and road network optimization. According to the invention, the obstacle is judged, and the judged obstacle is recorded or marked, so that in the subsequent road network updating, the marked area can be pertinently improved, the information can be accurately acquired, the misjudgment of the obstacle can be avoided, and on the basis of the judgment, the accuracy of the road network updating is improved. Therefore, errors in road network generation can be avoided to a certain extent or errors caused by obstacles in the road network can be corrected in time.

Description

Technical Field

[0001] The present invention relates to the technical field of driverless technology, and particularly to a method for generating a parking lot navigation road network for a driverless AVP scenario. Background Art

[0002] The parking lot navigation road network for a driverless AVP scenario is a road network and related information system specifically constructed for vehicles to achieve autonomous navigation and parking in a parking lot.

[0003] However, when generating the existing parking lot navigation road network for a driverless AVP scenario, since there are various movable obstacles in the parking lot, when generating the navigation road network of the parking lot, the obstacles will not only block the perception of the sensors, making the information of some areas unable to be accurately obtained. In addition, there may be a phenomenon of misjudging movable obstacles as fixed obstacles, which will lead to errors in the generation of the road network. Therefore, a method for generating a parking lot navigation road network for a driverless AVP scenario is invented. Summary of the Invention

[0004] To solve the above technical problems, according to one aspect of the present invention, the following technical solutions are provided:

[0005] A method for generating a parking lot navigation road network for a driverless AVP scenario, which includes the following specific steps:

[0006] S1, Data collection:

[0007] S11, Sensor deployment: Deploy a variety of sensors inside the parking lot, including lidar, cameras, and ultrasonic sensors; the lidar is used to obtain the three-dimensional point cloud data of the parking lot, accurately measure the positions and shapes of obstacles, parking spaces, and road boundaries; the cameras are used to capture the visual images of the parking lot, provide texture and color information, and assist in identifying lane lines, traffic signs, and other vehicles and pedestrians; the ultrasonic sensors are used to detect obstacles at close range; during the process of sensor deployment, if an obstacle is encountered, the obstacle will be analyzed to analyze whether the obstacle is a fixed-point obstacle or a moving-point obstacle. If it is a fixed-point obstacle, it will be recorded into the three-dimensional point cloud data. If it is a moving-point obstacle, it will be marked and the time will be recorded;

[0008] S12, Vehicle data collection: Make a collection vehicle equipped with a high-precision positioning system and an inertial measurement unit drive in the parking lot to collect the position, speed, and attitude information of the vehicle. The data is used to determine the driving trajectory and attitude changes of the vehicle, providing basic data for subsequent road network generation;

[0009] S13, Map Data Collection: Collect the electronic map data of the parking lot. The electronic map data can serve as the basic framework for road network generation, providing the general layout and geometric information of the parking lot;

[0010] S2, Data Processing and Analysis:

[0011] S21, Point Cloud Data Processing: First, perform filtering and denoising preprocessing operations on the point cloud data collected by the lidar to remove abnormal points and noise points, improving the quality and accuracy of the data. Then, perform point cloud segmentation to divide the point cloud data into different regions;

[0012] S22, Image Data Processing: Perform image recognition and analysis on the images collected by the camera to identify feature information. Then, use computer vision technology to extract useful information from the images and fuse it with the point cloud data to comprehensively understand the environment of the parking lot;

[0013] S23, Data Fusion: Use algorithms to fuse the data of lidar, camera, ultrasonic sensor, and the collection vehicle to obtain more accurate and comprehensive information about the parking lot environment;

[0014] S3, Road Network Generation:

[0015] S31, Node Determination: Based on the results of point cloud data and image data processing, determine the nodes of the parking lot navigation road network. The nodes will serve as the basic components of the road network, used to connect and construct navigation paths;

[0016] S32, Edge Connection: Determine the edges between nodes according to the actual physical connection relationship between nodes and the feasibility of vehicle driving. The edges represent the paths that vehicles can drive on and connect adjacent nodes; After determining the edges, the dimensions of the roads and parking spaces will be determined. After determination, it will be judged whether the vehicle can pass smoothly based on the vehicle size;

[0017] S33, Path Planning: Based on the generated road network, use path planning algorithms to plan the optimal navigation path for the vehicle from the starting point to the ending point;

[0018] S34, Road Network Optimization: Optimize the generated navigation road network, remove unnecessary nodes and edges, simplify the road network structure, improve the efficiency and accuracy of path planning. At the same time, optimize the positions of curves and intersections in the road network to ensure that the vehicle can pass smoothly and safely during driving;

[0019] S4, Map Update and Maintenance:

[0020] S41, New data update: The environment of the parking lot is monitored in real time according to data collection. During the monitoring process, if an obstacle appears in a space without obstacles, it will be analyzed through data collection to determine whether the obstacle is a fixed obstacle or a moving obstacle. If it is a moving obstacle, no update will be performed; if it is a fixed obstacle, an update will be performed.

[0021] S42, Old data update: According to the recording time in data collection, the marked area data is collected in sequence. If the moving obstacle has not moved, the time will be recorded again.

[0022] As a preferred solution of a method for generating a parking lot navigation road network for an unmanned AVP scenario according to the present invention, wherein: the electronic map data in S13 includes the floor plan, building structure, and entrance and exit positions of the parking lot.

[0023] As a preferred solution of a method for generating a parking lot navigation road network for an unmanned AVP scenario according to the present invention, wherein: the areas divided by the point cloud data in S21 include parking space areas, driving lane areas, and obstacle areas.

[0024] As a preferred solution of a method for generating a parking lot navigation road network for an unmanned AVP scenario according to the present invention, wherein: the feature information in S22 includes lane lines, traffic signs, and parking space markings, and the vision technology includes edge detection algorithms, feature extraction algorithms, and image classification algorithms.

[0025] As a preferred solution of a method for generating a parking lot navigation road network for an unmanned AVP scenario according to the present invention, wherein: the algorithms used in S23 include Kalman filter algorithms and particle filter algorithms.

[0026] As a preferred solution of a method for generating a parking lot navigation road network for an unmanned AVP scenario according to the present invention, wherein: the nodes in S31 include the entrances and exits of the parking lot, the centers of parking spaces, the intersections of roads, and the key points of curves.

[0027] As a preferred solution of a method for generating a parking lot navigation road network for an unmanned AVP scenario according to the present invention, wherein: the steps for determining the road size in S32 are as follows:

[0028] S3211, Image preprocessing:

[0029] S32111, Grayscale conversion: If the image is in color, convert it to a grayscale image to reduce the computational load and interference from color information.

[0030] S32112, Noise reduction: Use a filtering algorithm to perform noise reduction on the image to remove noise and interference in the image and improve the image quality;

[0031] S3212, Edge detection:

[0032] S32121, Selection algorithm: According to the characteristics and accuracy requirements of the image, select a suitable edge detection algorithm, and the algorithms include the Canny algorithm and the Sobel algorithm;

[0033] S32122, Application algorithm: Call the selected edge detection algorithm in an image processing software or programming language to perform edge detection on the preprocessed image. The algorithm will identify the edge of the road based on the gray-scale change of the pixels in the image and represent it in the form of pixel points;

[0034] S3213, Measure the road size:

[0035] S32131, Analyze the edge pixels: Analyze the detected road edge pixels and calculate the distance between the edge pixel points. By counting the number of adjacent edge pixel points and combining the resolution of the image, convert the pixel distance into an actual size.

[0036] As a preferred solution of a method for generating a parking lot navigation road network for an unmanned AVP scenario according to the present invention, wherein: the steps for determining the size of the parking space in S32 are as follows:

[0037] S3221, Image acquisition: Simultaneously capture scenes containing parking spaces from different angles to obtain multiple images, ensuring that there is sufficient overlapping area of the parking spaces in different images;

[0038] S3222, Stereo matching: Perform stereo matching on the obtained multiple images to find the corresponding points of the same object in different images;

[0039] S3223, Depth calculation: According to the corresponding points obtained by stereo matching, use the principle of triangulation to calculate the depth information of each point, that is, the distance between the point and the camera;

[0040] S3224, Size calculation:

[0041] S32241: According to the depth information and the coordinate information in the image, convert the points on the parking space into a three-dimensional space coordinate system;

[0042] S32242: Calculate the distances and angles between the vertices of the parking space in the three-dimensional space, so as to obtain the actual shape and size of the parking space.

[0043] Compared with the prior art:

[0044] By judging obstacles and entering or marking the judged obstacles, the present invention can not only specifically improve the marked area in subsequent road network updates to accurately obtain information, but also avoid misjudging obstacles. Based on this, it can to a certain extent avoid errors in road network generation or promptly correct errors in the road network caused by obstacles. In addition, by determining the sizes of roads and parking spaces, it can judge whether vehicles can pass smoothly on roads and in parking spaces. Detailed implementation manners

[0045] To make the objectives, technical solutions and advantages of the present invention clearer, the following will further describe the implementation manners of the present invention in detail.

[0046] The present invention provides a method for generating a parking lot navigation road network for an unmanned AVP scenario, including the following specific steps:

[0047] S1, Data collection:

[0048] S11, Sensor deployment: Deploy a variety of sensors inside the parking lot, including lidar, cameras, and ultrasonic sensors; the lidar is used to obtain the three-dimensional point cloud data of the parking lot and accurately measure the positions and shapes of obstacles, parking spaces, and road boundaries; the cameras are used to capture the visual images of the parking lot, provide texture and color information, and assist in identifying lane lines, traffic signs, and other vehicles and pedestrians; the ultrasonic sensors are used to detect obstacles at close range; during the sensor deployment process, if an obstacle is encountered, the obstacle will be analyzed to determine whether it is a fixed-point obstacle or a moving-point obstacle. If it is a fixed-point obstacle, it will be entered into the three-dimensional point cloud data. If it is a moving-point obstacle, it will be marked and the time will be recorded.

[0049] S12, Vehicle data collection: Let the collection vehicle equipped with a high-precision positioning system and an inertial measurement unit drive inside the parking lot to collect the position, speed, and attitude information of the vehicle. The data is used to determine the driving trajectory and attitude changes of the vehicle and provide basic data for subsequent road network generation.

[0050] S13, Map data collection: Collect the electronic map data of the parking lot. The electronic map data includes the floor plan, building structure, and entrance and exit positions of the parking lot. The electronic map data can serve as the basic framework for road network generation and provide the general layout and geometric information of the parking lot.

[0051] S2, Data processing and analysis:

[0052] S21, Point cloud data processing: First, perform filtering and denoising preprocessing operations on the point cloud data collected by the lidar to remove abnormal points and noise points, improve the quality and accuracy of the data, and then perform point cloud segmentation to divide the point cloud data into different regions. Among them, the regions where the point cloud data is divided include parking space regions, driving lane regions, and obstacle regions;

[0053] S22, Image data processing: Perform image recognition and analysis on the images collected by the camera to identify feature information, including lane lines, traffic signs, and parking space markings. Then, use computer vision technology to extract useful information from the images and fuse it with the point cloud data to more comprehensively understand the parking lot environment. The vision technology includes edge detection algorithms, feature extraction algorithms, and image classification algorithms;

[0054] S23, Data fusion: Use algorithms (including Kalman filtering algorithm and particle filtering algorithm) to fuse the data of lidar, camera, ultrasonic sensor, and the collection vehicle to obtain more accurate and comprehensive parking lot environment information;

[0055] S3, Road network generation:

[0056] S31, Node determination: According to the results of point cloud data and image data processing, determine the nodes of the parking lot navigation road network. The nodes include the entrances and exits of the parking lot, the centers of parking spaces, the intersections of roads, and the key points of curves. These nodes will serve as the basic units of the road network for connecting and constructing navigation paths;

[0057] S32, Edge connection: Determine the edges between nodes according to the actual physical connection relationship between nodes and the feasibility of vehicle driving. The edges represent the paths that vehicles can drive and connect adjacent nodes; After determining the edges, the sizes of the roads and parking spaces will be determined. After determination, it will be judged whether the vehicle can pass smoothly based on the vehicle size;

[0058] Among them:

[0059] The steps for determining the road size are as follows:

[0060] S3211, Image preprocessing:

[0061] S32111, Grayscale conversion: If the image is in color, convert it to a grayscale image to reduce the amount of calculation and interference from color information;

[0062] S32112, Denoising: Use a filtering algorithm to perform denoising processing on the image to remove noise and interference in the image and improve the image quality;

[0063] S3212, Edge detection:

[0064] S32121, Selection Algorithm: According to the characteristics of the image and the accuracy requirements, select an appropriate edge detection algorithm, including the Canny algorithm and the Sobel algorithm;

[0065] S32122, Algorithm Application: Call the selected edge detection algorithm in an image processing software or programming language to perform edge detection on the preprocessed image. The algorithm will identify the edges of the road based on the gray-scale changes of the pixels in the image and represent them in the form of pixel points;

[0066] S3213, Measuring Road Dimensions:

[0067] S32131, Analyzing Edge Pixels: Analyze the detected road edge pixels and calculate the distances between the edge pixel points. By counting the number of adjacent edge pixel points and combining with the resolution of the image, convert the pixel distances into actual dimensions;

[0068] The steps for determining the parking space dimensions are as follows:

[0069] S3221, Image Acquisition: Simultaneously capture scenes containing the parking space from different angles to obtain multiple images, ensuring that there is sufficient overlapping area of the parking space in different images;

[0070] S3222, Stereo Matching: Perform stereo matching on the obtained multiple images to find the corresponding points of the same object in different images;

[0071] S3223, Depth Calculation: Based on the corresponding points obtained from stereo matching, use the principle of triangulation to calculate the depth information of each point, i.e., the distance between the point and the camera;

[0072] S3224, Dimension Calculation:

[0073] S32241: Convert the points on the parking space into a three-dimensional space coordinate system based on the depth information and the coordinate information in the image;

[0074] S32242: Calculate the distances and angles between the vertices of the parking space in three-dimensional space to obtain the actual shape and size of the parking space. The steps for determining the parking space dimensions are as follows:

[0075] S3221, Image Acquisition: Simultaneously capture scenes containing the parking space from different angles to obtain multiple images, ensuring that there is sufficient overlapping area of the parking space in different images;

[0076] S3222, Stereo Matching: Perform stereo matching on the obtained multiple images to find the corresponding points of the same object in different images;

[0077] S3223, Depth calculation: Based on the corresponding points obtained from stereo matching, the depth information of each point, that is, the distance between the point and the camera, is calculated using the principle of triangulation;

[0078] S3224, Dimension calculation:

[0079] S32241: According to the depth information and the coordinate information in the image, the points on the parking space are transformed into a three-dimensional space coordinate system;

[0080] S32242: Calculate the distances and angles between the vertices of the parking space in three-dimensional space, so as to obtain the actual shape and size of the parking space;

[0081] S33, Path planning: Based on the generated road network, a path planning algorithm is used to plan the optimal navigation path for the vehicle from the starting point to the ending point;

[0082] S34, Road network optimization: Optimize the generated navigation road network, remove unnecessary nodes and edges, simplify the road network structure, improve the efficiency and accuracy of path planning. At the same time, optimize the positions of curves and intersections in the road network to ensure that the vehicle can pass smoothly and safely during driving;

[0083] S4, Map update and maintenance:

[0084] S41, New data update: According to data collection, the environment of the parking lot is monitored in real time. During the monitoring process, if an obstacle appears in a space without obstacles, it will be analyzed through data collection to determine whether the obstacle is a fixed obstacle or a moving obstacle. If it is a moving obstacle, no update will be performed. If it is a fixed obstacle, an update will be performed;

[0085] S42, Old data update: According to the recording time in data collection, the marked area data is collected in sequence. If the moving obstacle has not moved, the time will be recorded again.

[0086] Although the present invention has been described above with reference to the embodiments, various improvements can be made to it and its components can be replaced with equivalents without departing from the scope of the present invention. In particular, as long as there is no structural conflict, the features in the disclosed embodiments of the present invention can be combined with each other in any way. The exhaustive description of these combinations is not given in this specification only for the sake of saving space and resources. Therefore, the present invention is not limited to the specific embodiments disclosed in the text, but includes all technical solutions falling within the scope of the claims.

Claims

1. A method for generating a parking lot navigation network for an unmanned AVP scenario, characterized in that: The specific steps are as follows: S1, Data Collection: S11, Sensor deployment: Various sensors are deployed in the parking lot, including LiDAR, cameras, and ultrasonic sensors; LiDAR is used to obtain 3D point cloud data of the parking lot and accurately measure the position and shape of obstacles, parking spaces, and road boundaries; cameras are used to capture visual images of the parking lot, provide texture and color information, and assist in identifying lane lines, traffic signs, and other vehicles and pedestrians; ultrasonic sensors are used to detect obstacles at close range; during the sensor deployment process, if an obstacle is encountered, the obstacle will be analyzed to determine whether it is a fixed-point obstacle or a moving-point obstacle. If it is a fixed-point obstacle, it will be recorded in the 3D point cloud data; if it is a moving-point obstacle, it will be marked and the time will be recorded; S12, vehicle data collection: a collection vehicle equipped with a high-precision positioning system and an inertial measurement unit is driven in the parking lot to collect the position, speed, and posture information of the vehicle. The data is used to determine the driving trajectory and posture changes of the vehicle, and provide basic data for subsequent road network generation; S13, map data collection: collecting electronic map data of the parking lot, which can be used as a basic framework for road network generation and provide a general layout and geometric information of the parking lot; S2, Data processing and analysis: S21, point cloud data processing: first, filter and denoise the point cloud data collected by the lidar to remove abnormal points and noise points, improve the quality and accuracy of the data, and then perform point cloud segmentation to divide the point cloud data into different areas; S22, image data processing: image recognition and analysis are performed on the images collected by the camera to identify feature information, and then useful information is extracted from the images using computer vision technology and integrated with point cloud data to more comprehensively understand the parking environment; S23, data fusion: use algorithms to fuse data from lidar, cameras, ultrasonic sensors and collection vehicles to obtain more accurate and comprehensive parking environment information; S3, road network generation: S31, node determination: determining the nodes of the parking lot navigation network according to the results of point cloud data and image data processing, and the nodes will be used as the basic components of the network to connect and construct the navigation path; S32, edge connection: determine the edges between nodes according to the actual physical connection relationship between nodes and the feasibility of vehicle driving, where the edges represent the paths that vehicles can drive and connect adjacent nodes; after determining the edges, the size of the road and the size of the parking space will be determined, and after determination, it will be judged whether the vehicle can pass smoothly based on the vehicle size; S33, path planning: based on the generated road network, a path planning algorithm is used to plan an optimal navigation path from the starting point to the end point for the vehicle; S34, road network optimization: optimize the generated navigation road network, remove unnecessary nodes and edges, simplify the road network structure, improve the efficiency and accuracy of path planning, and optimize the positions of curves and intersections in the road network to ensure that the vehicle can pass smoothly and safely during driving; S4, map update and maintenance: S41, new data update: the parking lot environment is monitored in real time according to data collection. During the monitoring process, if an obstacle appears in the obstacle-free space, data collection is used to analyze whether the obstacle is a fixed-point obstacle or a moving-point obstacle. If it is a moving-point obstacle, no update will be performed; if it is a fixed-point obstacle, an update will be performed; S42, old data update: according to the recording time in data collection, the marked area data is collected in sequence. If the moving point obstacle has not moved, the time will be recorded again.

2. According to the method for generating a parking lot navigation network for an unmanned driving AVP scenario according to claim 1, it is characterized in that: The electronic map data in S13 includes the floor plan, building structure, and entrance and exit locations of the parking lot.

3. According to the method for generating a parking lot navigation network for an unmanned driving AVP scenario according to claim 1, it is characterized in that: The areas in which the point cloud data is divided in S21 include a parking space area, a driving lane area, and an obstacle area.

4. According to the method for generating a parking lot navigation network for an unmanned driving AVP scenario according to claim 1, it is characterized in that: The feature information in S22 includes lane lines, traffic signs, and parking space markings, and the visual technology includes edge detection algorithms, feature extraction algorithms, and image classification algorithms.

5. The method for generating a parking lot navigation network for an unmanned driving AVP scenario according to claim 1 is characterized in that: The algorithms used in S23 include a Kalman filter algorithm and a particle filter algorithm.

6. A method for generating a parking lot navigation network for an unmanned driving AVP scenario according to claim 1, characterized in that: The nodes in S31 include the entrance and exit of the parking lot, the center of the parking space, the intersection of the road, and the key points of the curve.

7. A method for generating a parking lot navigation network for an unmanned driving AVP scenario according to claim 1, characterized in that: The steps for determining the road size in S32 are as follows: S3211, image preprocessing: S32111, Grayscale: If the image is in color, convert it to a grayscale image to reduce the amount of calculation and the interference of color information; S32112, Noise Reduction: Use filtering algorithms to reduce the noise of images to remove noise and interference in the images and improve image quality; S3212, edge detection: S32121, select algorithm: select appropriate edge detection algorithm according to the characteristics and accuracy requirements of the image, including Canny algorithm and Sobel algorithm; S32122, application algorithm: calling the selected edge detection algorithm in the image processing software or programming language to perform edge detection on the preprocessed image. The algorithm will identify the edge of the road according to the grayscale change of the pixels in the image and represent it in the form of pixel points; S3213, Measuring road dimensions: S32131, analyze edge pixels: analyze the detected road edge pixels and calculate the distance between edge pixels, so as to convert the pixel distance into actual size by counting the number of adjacent edge pixels and combining the resolution of the image.

8. The method for generating a parking lot navigation network for an unmanned driving AVP scenario according to claim 1, characterized in that: The steps for determining the parking space size in S32 are as follows: S3221, image acquisition: simultaneously photographing a scene including a parking space from different angles to obtain multiple images, ensuring that the parking space has sufficient overlapping areas in different images; S3222, stereo matching: performing stereo matching on the acquired multiple images to find corresponding points of the same object in different images; S3223, depth calculation: according to the corresponding points obtained by stereo matching, the depth information of each point is calculated using the triangulation principle, that is, the distance between the point and the camera; S3224, Dimension Calculation: S32241: Converting points on the parking space into a three-dimensional space coordinate system according to the depth information and the coordinate information in the image; S32242: Calculate the distance and angle between the vertices of the parking space in three-dimensional space to obtain the actual shape and size of the parking space.