Method for constructing stereo garage map in navigation system

Abstract

The present invention discloses a method for constructing a stereoscopic parking garage map in a navigation system. The method comprises the following steps: calibrating a camera, a laser radar, and an inertial measurement unit of a vehicle, obtaining the vehicle's posture transformation data through the inertial measurement unit, and obtaining an estimated posture of the vehicle in the stereoscopic parking garage map after correction; obtaining the vehicle's environmental point cloud data through the laser radar, and converting it into a point cloud sub-graph; matching the point cloud sub-graph at a current moment with the point cloud sub-graph at a previous moment, and obtaining a matching posture based on the laser radar; obtaining image data and depth data in front of the vehicle, segmenting the data, and projecting the data into a point cloud sub-graph with the same timestamp, and obtaining a fused point cloud sub-graph with multiple sensors; stripping the ground point cloud in the fused point cloud sub-graph according to the point cloud height, obtaining a ground point cloud of only parking space lines, and then updating the data into the fused point cloud sub-graph; supplementing the updated fused point cloud sub-graph in the vehicle's navigation system map, and constructing a map of the stereoscopic parking garage.

Description

Technical Field

[0001] The present invention relates to the technical field of automatic navigation, and in particular to a method for constructing a stereoscopic parking garage map in a navigation system. Background Art

[0002] Real-time positioning and mapping are key steps in autonomous navigation technology. Using sensors to gather environmental information in unexplored environments, the vehicle can determine its position within the environment and construct a map of the surroundings. This technology can also identify the distance between the vehicle and obstacles, aiding in path planning and obstacle avoidance. When used in a parking garage, this technology allows vehicle transfer carts to follow planned routes, enabling convenient pickup and parking without the need for human drivers. However, existing methods for real-time positioning and mapping for vehicle transfer carts have limitations, including limited reliability in positioning and mapping.

[0003] The invention patent with publication number CN112627611A discloses an intelligent AGV handling robot and its handling method, which uses laser SLAM and a QR code scanner: the laser SLAM system is used to locate its own posture, and the core control unit drives the motor to the predetermined position and adjusts its own posture according to the posture control motion module. The QR code memory at the predetermined position stores the posture information of the previous stage path and the end point of the path, which is read by the QR code scanner and handed over to the core control unit to parse the content and calculate the current posture deviation. After that, it enters the secondary positioning state to calculate and obtain the coordinates of the fine-tuned corner points, and then plans a new curved path based on the original path, and finally returns to the path movement process positioned by laser SLAM.

[0004] The positioning method in this solution is limited, relying solely on laser SLAM to estimate pose without cross-validating with an inertial measurement unit. Secondly, the path information stored in the QR code is fixed, and subsequent planning modifications are made based on this path. This can easily lead to path overlap when multiple vehicles converge. If a preceding vehicle incorrectly parks, preventing the following vehicle from reading the QR code, the entire garage will be unable to operate, limiting its automated intelligence. Furthermore, the laser SLAM method only captures obstacle features at radar height, calculates distance using time-of-flight methods, and uses line features instead of plane features to create a two-dimensional map. A significant amount of spatial information and features is not considered or calculated. Finally, changes in the QR code position and information within the code due to environmental changes require manual resetting, which is somewhat inefficient in terms of adaptability and timeliness. Therefore, further consideration is needed to provide a highly accurate method for constructing a three-dimensional garage map that includes ground-level parking space line information. Summary of the Invention

[0005] In view of the above-mentioned deficiencies in the prior art, the technical problem to be solved by the present invention is: how to provide a method for constructing a stereoscopic parking garage map in a navigation system that includes ground parking space line information and has high positioning accuracy.

[0006] In order to solve the above technical problems, the present invention adopts the following technical solutions:

[0007] A method for constructing a stereoscopic parking garage map in a navigation system comprises the following steps:

[0008] (1) Calibrate the sensors on the car with automatic navigation function. The sensors include cameras, laser radars, and inertial measurement units. The camera can obtain images in front of the car. The camera's intrinsic parameter matrix and extrinsic parameter matrix, as well as the joint data conversion matrix of the laser radar and the inertial measurement unit are obtained through sensor calibration. The intrinsic parameter matrix and extrinsic parameter matrix of each camera are different. The specific parameters are determined through calibration. The purpose is to ensure accuracy when converting camera data to fused point cloud data. The measurement effect of each laser radar is different. The joint data conversion matrix obtained through calibration can compensate the measurement value to make it close to the true value, which is better when converting laser radar data to point cloud sub-image data.

[0009] (2) The posture change data of the car is obtained through the inertial measurement unit, and the posture change data is processed by the extended Kalman filter to obtain the estimated posture of the car in the stereo garage map after correction; the estimated posture is the initial value for the subsequent point cloud sub-map matching work. The estimated posture after the extended Kalman filter processing is more accurate than the original posture information output by the inertial measurement unit. The estimated posture after the extended Kalman filter processing has less noise information and less posture offset;

[0010] (3) Obtain the environmental point cloud data of the car through the laser radar, convert the environmental point cloud data into a point cloud sub-graph according to the joint data conversion matrix, use the estimated pose as the initial value and use the point cloud sub-graph at this moment to match the point cloud sub-graph at the previous moment to obtain the matching pose based on the laser radar; Based on the estimated pose, use the information of the laser radar to correct the pose, and obtain the matching pose that is the value closest to the true pose. The navigation system area obtained based on the matching pose is more accurate;

[0011] (4) After acquiring the image data and depth data in front of the car through the camera, the image data and depth data are segmented using the intrinsic parameter matrix and the extrinsic parameter matrix and projected into the point cloud sub-image of the same timestamp to obtain a fused point cloud sub-image with multiple sensors. This step adds the camera data to the point cloud sub-image to form a multi-sensor fused point cloud sub-image, so that the fused point cloud sub-image contains more features.

[0012] (5) Preprocess the fused point cloud sub-map, peel off the ground point cloud in the fused point cloud sub-map according to the point cloud height, perform image processing on the ground point cloud to obtain a ground point cloud with only parking space lines, and then update it to the fused point cloud sub-map; the parking space line information of the stereo garage is an important part of the map information, and this part of the features is extracted and saved through image processing, so that the updated fused point cloud sub-map has richer information;

[0013] (6) The updated fused point cloud subgraph is added to the navigation system map of the car based on the matching posture of the lidar to construct a map of the stereoscopic parking garage; the navigation system map is constructed using the fused point cloud subgraph, which includes all the above-mentioned sensor information, and uses the matching posture as a reference, which is closer to the appearance of the real map and minimizes the error as much as possible.

[0014] As an optimization, in step (1), the camera calibration work includes using the CameraCalibrator in the matlab toolbox to perform Zhang Zhengyou calibration, and obtaining the camera's intrinsic parameter matrix and extrinsic parameter matrix in the cameraParameter attribute. The intrinsic parameter matrix includes the focal length information f x and f y , main point information c x and c y , the external parameter matrix includes the rotation information matrix R and the translation matrix t;

[0015] The calibration of the laser radar and the inertial measurement unit includes using the lidar-align function package for calibration. After inputting the posture transformation data obtained by the inertial measurement unit and the environmental point cloud data obtained by the laser radar, the least squares method is used to calculate the transformation matrix that minimizes the average distance.

[0016] As an optimization, in step (2), the extended Kalman filter method includes:

[0017] According to the position p of the car at the previous moment t-1 and variance ε t-1 The pose p estimated by the inertial measurement unit at that moment t and variance ε t Calculate the Kalman gain matrix k t , use Kalman gain to update the pose at that moment and variance The calculation formula is:

[0018]

[0019] where g(u t ,p t-1 ) is the state transition probability, G t and Ht is the Jacobian matrix, Q t and R t is Gaussian noise, h is the measurement probability, and I is the identity matrix.

[0020] As an optimization, in step (3), the point cloud sub-image conversion method includes:

[0021] First, the laser radar sampling data z t =(l,θ), l is the sampling distance, θ is the sampling angle, converted to Euclidean coordinates (x t ,y t ) is:

[0022]

[0023] Set the pose to ξ = (x0; y0; θ0), and set the initial value to the pose calculated by the Kalman gain in step (2) Together with the Euclidean coordinate values ​​of the environmental point cloud data, they are spliced ​​into a data matrix in the following format: The superscript of the character indicates the order of the value in a single measurement cycle, and n represents the sampling frequency of the sampling cycle. The LiDAR matching pose is obtained for subsequent matching work.

[0024] As an optimization, in step (3), the matching method of the point cloud sub-image at this moment and the point cloud sub-image at the previous moment includes:

[0025] Use the Scan-Match method to find the optimal (x0, y0, θ0) so that the point cloud sub-image at this moment and the point cloud sub-image at the previous moment can achieve the maximum probability of feature point overlap, by calculating the minimum constraint:

[0026]

[0027] Among them, M smooth is the data after smoothing. It is p t The i-th radar adopts data under the pose, ξ is the above conversion formula (4), and the estimated pose obtained in step (2) is calculated. Substitute the initial value into formula (5) to reduce the amount of calculation, use the least squares method to calculate the point cloud sub-image at this moment and the point cloud sub-image at the previous moment in pairs, and finally obtain the matching pose (x0, y0, θ0) of the lidar.

[0028] As an optimization, in step (4), the generation method of the fused point cloud subgraph includes:

[0029] Use the depthimage_to_laserscan function in the ROS package to divide the image data acquired by the camera into sub-images of equal size. The depth information of the sub-image is converted into the length from the focal point. The position of the sub-image in the point cloud sub-image is calculated by the intrinsic parameter matrix and the extrinsic parameter matrix. The calculation formula of the camera coordinate point (u, v) and the spatial coordinate point (X, Y, Z) is:

[0030]

[0031] where f x and f y is the focal length information, c x and c y is the main point information, R is the rotation information matrix, and t is the translation matrix.

[0032] As an optimization, in step (5), the ground point cloud is stripped in the following manner:

[0033] Set the ground height threshold to d, and place all points in the fused point cloud sub-graph whose absolute value of Z-axis height is less than d in set S. Set S is the candidate set of ground information features. Calculate the smoothness of the points in set S. After removing outliers, set S can be considered as the stripped ground information point cloud.

[0034] As an optimization, in step (5), the ground point cloud is stripped in the following manner:

[0035] The key frames of the image data acquired by the camera are extracted, and then the RGB image is converted into an HSV image. After Gaussian filtering and image enhancement, it is converted back into an RGB image to improve image differentiation. The image is then grayscaled and binarized to highlight the edge information of the parking space lines in the image. After that, connected area marking is used to strengthen the connection between the edge information. Then, edge detection is used to extract the contour information of the parking space lines. Finally, dilation and erosion processing is used to construct the complete parking space line information based on the contour.

[0036] As an optimization, in step (6), the method of using the updated fused point cloud subgraph to supplement the navigation system map of the car includes:

[0037] A linked list is set to save the point cloud subgraph and the matching pose at each moment. When the consecutive frames after a certain moment successfully complete the matching of step (3), the dependency between the previous and next frames is realized, and the pose is corrected according to the time difference between the key frames.

[0038] In summary, the beneficial effects of the present invention are:

[0039] 1. Using depth cameras and lidar as environmental information sensors, the lidar obtains 360-degree map information at the same height, and the depth camera obtains image information directly in front of the vehicle. This complements the advantages of the lidar (which has a wide range but partial information) and the depth camera (which has a large information volume but a small range). The fused point cloud sub-image obtained by the two is more detailed than that obtained by a single sensor, and the constructed stereo garage map also has more feature points, which facilitates subsequent positioning and navigation work and improves reliability in stereo garages.

[0040] 2. Use keyframe dependency to sort the car's pose and fused point cloud sub-images in the previous and next frames. During loop closure detection, the last frame is aligned with the previous frame. Then, each keyframe between the two frames is corrected from back to front to achieve perfect loop closure.

[0041] 3. The pose change data recorded by the inertial measurement unit is processed using an extended Kalman filter. There is a random walk between the sensor's detected value and the actual value, and the wandering value of a frame is passed to the next frame as a keyframe dependency. This error accumulation phenomenon needs to be reduced using filtering methods. At the same time, the sensor does not return a zero value when no data value is detected, but varies within a certain threshold close to zero. This zero bias phenomenon is also reduced during the filtering process.

[0042] 4. The series of image processing technologies provided can extract parking space line information on the ground, eliminate other irrelevant information from the ground content, and only retain the parking space line information required for subsequent parking and pick-up functions of small vehicles, making the information clear, independent, and the map concise. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] In order to make the purpose, technical solutions and advantages of the invention more clear, the present invention will be further described in detail below with reference to the accompanying drawings, in which:

[0044] Figure 1 is a flow chart of the present invention;

[0045] Figure 2 Flowchart of image processing in the present invention. DETAILED DESCRIPTION

[0046] like Figure 1 and Figure 2 As shown, the method for constructing a stereoscopic parking garage map in the navigation system in this specific embodiment includes the following steps:

[0047] (1) Calibrate the sensors on the car with automatic navigation function. The sensors include a camera, a lidar, and an inertial measurement unit. The camera can obtain the image in front of the car. The intrinsic and extrinsic parameter matrices of the camera, as well as the joint data conversion matrix of the lidar and the inertial measurement unit, are obtained through the calibration of the sensors.

[0048] (2) Obtaining the car's posture change data through the inertial measurement unit, performing extended Kalman filtering on the posture change data to obtain the estimated posture of the car in the stereo garage map after correction;

[0049] (3) Obtain the environmental point cloud data of the car through the laser radar, convert the environmental point cloud data into a point cloud sub-image according to the joint data conversion matrix, use the estimated pose as the initial value and use the point cloud sub-image at this moment to match the point cloud sub-image at the previous moment to obtain the matching pose based on the laser radar;

[0050] (4) After acquiring the image data and depth data in front of the car through the camera, the image data and depth data are segmented using the intrinsic parameter matrix and the extrinsic parameter matrix and projected into the point cloud sub-image of the same timestamp to obtain the fused point cloud sub-image with multiple sensors;

[0051] (5) Preprocess the fused point cloud sub-image, peel off the ground point cloud in the fused point cloud sub-image according to the point cloud height, perform image processing on the ground point cloud to obtain the ground point cloud of only the parking space line, and then update it into the fused point cloud sub-image;

[0052] (6) The updated fused point cloud sub-graph is added to the navigation system map of the car based on the matching posture of the lidar to construct the map of the stereo garage.

[0053] In this specific implementation, in step (1), the camera calibration work includes using the CameraCalibrator in the matlab toolbox to perform Zhang Zhengyou calibration, and obtaining the camera's intrinsic parameter matrix and extrinsic parameter matrix in the cameraParameter attribute. The intrinsic parameter matrix includes focal length information f x and f y , main point information c x and c y , the external parameter matrix includes the rotation information matrix R and the translation matrix t;

[0054] The calibration of the laser radar and the inertial measurement unit includes using the lidar-align function package for calibration. After inputting the posture transformation data obtained by the inertial measurement unit and the environmental point cloud data obtained by the laser radar, the least squares method is used to calculate the transformation matrix that minimizes the average distance.

[0055] In this specific implementation, in step (2), the extended Kalman filter method includes:

[0056] According to the position p of the car at the previous moment t-1 and variance ε t-1 The pose p estimated by the inertial measurement unit at that moment tand variance ε t Calculate the Kalman gain matrix k t , use Kalman gain to update the pose at that moment and variance The calculation formula is:

[0057]

[0058] where g(u t ,p t-1 ) is the state transition probability, G t and H t is the Jacobian matrix, Q t and R t is Gaussian noise, h is the measurement probability, and I is the identity matrix.

[0059] In this specific implementation, in step (3), the point cloud sub-graph conversion method includes:

[0060] First, the laser radar sampling data z t =(l,θ), l is the sampling distance, θ is the sampling angle, converted to Euclidean coordinates (x t ,y t ), the conversion formula is:

[0061]

[0062] Set the pose to ξ = (x0, y0, θ0), and set the initial value to the pose calculated by the Kalman gain in step (2) Together with the Euclidean coordinate values ​​of the environmental point cloud data, they are spliced ​​into a data matrix in the following format: The superscript of the character indicates the order of the value in a single measurement cycle, and n indicates the sampling frequency of the sampling cycle.

[0063] In this specific implementation, in step (3), the point cloud sub-image at this moment and the point cloud sub-image at the previous moment are matched in the following manner:

[0064] Use the Scan-Match method to find the optimal (x0, y0, θ0) so that the point cloud sub-image at this moment and the point cloud sub-image at the previous moment can achieve the maximum probability of feature point overlap, by calculating the minimum constraint:

[0065]

[0066] Among them, M smooth is the data after smoothing. It is p t The i-th radar adopts data under the pose, ξ is the above conversion formula (4), and the estimated pose obtained in step (2) is calculated. As the initial value, it is substituted into formula (5) to reduce the amount of calculation. The point cloud sub-image at this moment and the point cloud sub-image at the previous moment are calculated pairwise using the least squares method, and finally the matching pose (x0, y0, θ0) of the laser radar is obtained.

[0067] In this specific implementation, in step (4), the generation method of the fused point cloud subgraph includes:

[0068] Use the depthimage_to_laserscan function in the ROS package to divide the image data acquired by the camera into sub-images of equal size. The depth information of the sub-image is converted into the length from the focal point. The position of the sub-image in the point cloud sub-image is calculated by the intrinsic parameter matrix and the extrinsic parameter matrix. The calculation formula of the camera coordinate point (u, v) and the spatial coordinate point (X, Y, Z) is:

[0069]

[0070] where f x and f y is the focal length information, c x and c y is the main point information, R is the rotation information matrix, and t is the translation matrix.

[0071] In this specific implementation, in step (5), the ground point cloud is stripped in the following manner:

[0072] Set the ground height threshold to d, and place all points in the fused point cloud sub-graph whose absolute value of Z-axis height is less than d in set S. Set S is the candidate set of ground information features. Calculate the smoothness of the points in set S. After removing outliers, set S can be considered as the stripped ground information point cloud.

[0073] In this specific implementation, in step (5), the ground point cloud is stripped in the following manner:

[0074] The key frames of the image data acquired by the camera are extracted, and then the RGB image is converted into an HSV image. After Gaussian filtering and image enhancement, it is converted back into an RGB image to improve image differentiation. The image is then grayscaled and binarized to highlight the edge information of the parking space lines in the image. After that, connected area marking is used to strengthen the connection between the edge information. Then, edge detection is used to extract the contour information of the parking space lines. Finally, dilation and erosion processing is used to construct the complete parking space line information based on the contour.

[0075] In this specific implementation, in step (6), the method of using the updated fused point cloud subgraph to supplement the navigation system map of the vehicle includes:

[0076] A linked list is set to save the point cloud subgraph and the matching pose at each moment. When the consecutive frames after a certain moment successfully complete the matching of step (3), the dependency between the previous and next frames is realized, and the pose is corrected according to the time difference between the key frames.

[0077] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not limiting. Although the present invention has been described with reference to the preferred embodiments of the present invention, it should be understood by those skilled in the art that various changes can be made in form and details without departing from the spirit and scope of the present invention as defined in the appended claims.

Claims

1. A method for constructing a stereoscopic parking garage map in a navigation system, characterized by: The following steps are involved: (1) Calibrate the sensors on the car with automatic navigation function. The sensors include a camera, a lidar, and an inertial measurement unit. The camera can obtain the image in front of the car. The intrinsic and extrinsic parameter matrices of the camera, as well as the joint data conversion matrix of the lidar and the inertial measurement unit, are obtained through the calibration of the sensors. (2) Obtaining the car's posture change data through the inertial measurement unit, performing extended Kalman filtering on the posture change data to obtain the estimated posture of the car in the stereo garage map after correction; The extended Kalman filter method includes: According to the position p of the car at the previous moment t-1 and variance ε t-1 The pose p estimated by the inertial measurement unit at that moment t and variance ε t Calculate the Kalman gain matrix k t , use Kalman gain to update the pose at that moment and variance The calculation formula is: where g(u t ,p t-1 ) is the state transition probability, G t and H t is the Jacobian matrix, Q t and R t is Gaussian noise, h is the measurement probability, and I is the identity matrix; (3) Obtain the environmental point cloud data of the car through the laser radar, convert the environmental point cloud data into a point cloud sub-image according to the joint data conversion matrix, use the estimated pose as the initial value and use the point cloud sub-image at this moment to match the point cloud sub-image at the previous moment to obtain the matching pose based on the laser radar; (4) After acquiring the image data and depth data in front of the car through the camera, the image data and depth data are segmented using the intrinsic parameter matrix and the extrinsic parameter matrix and projected into the point cloud sub-image of the same timestamp to obtain the fused point cloud sub-image with multiple sensors; (5) Preprocess the fused point cloud sub-image, peel off the ground point cloud in the fused point cloud sub-image according to the point cloud height, perform image processing on the ground point cloud to obtain the ground point cloud of only the parking space line, and then update it into the fused point cloud sub-image; (6) The updated fused point cloud sub-graph is added to the navigation system map of the car based on the matching posture of the lidar to construct the map of the stereo garage.

2. The method for constructing a stereoscopic parking garage map in a navigation system according to claim 1, characterized in that: In step (1), the camera calibration work includes using the CameraCalibrator in the matlab toolbox to perform Zhang Zhengyou calibration, and obtaining the camera's intrinsic parameter matrix and extrinsic parameter matrix in the cameraParameter attribute. The intrinsic parameter matrix includes the focal length information f x and f y , main point information c x and c y , the external parameter matrix includes the rotation information matrix R and the translation matrix t; The calibration of the laser radar and the inertial measurement unit includes using the lidar-align function package for calibration. After inputting the posture transformation data obtained by the inertial measurement unit and the environmental point cloud data obtained by the laser radar, the least squares method is used to calculate the transformation matrix that minimizes the average distance.

3. The method for constructing a stereoscopic parking garage map in a navigation system according to claim 1, characterized in that: In step (3), the point cloud sub-image conversion method includes: First, the laser radar sampling data z t =(l,θ), l is the sampling distance, θ is the sampling angle, converted to Euclidean coordinates (x t ,y t ) is: Set the pose to ξ = (x0, y0, θ0), and set the initial value to the pose calculated by the Kalman gain in step (2) Together with the Euclidean coordinate values ​​of the environmental point cloud data, they are spliced ​​into a data matrix in the following format: The superscript of the character indicates the order of the value in a single measurement cycle, and n indicates the sampling frequency of the sampling cycle.

4. The method for constructing a stereoscopic parking garage map in a navigation system according to claim 3, wherein: In step (3), the point cloud sub-image at this moment and the point cloud sub-image at the previous moment are matched in the following manner: Use the Scan-Match method to find the optimal (x0, y0, θ0) so that the point cloud sub-image at this moment and the point cloud sub-image at the previous moment can achieve the maximum probability of feature point overlap, by calculating the minimum constraint: Among them, M smooth is the data after smoothing. It is p t The i-th radar adopts data under the pose, ξ is the above conversion formula (4), and the estimated pose obtained in step (2) is calculated. Substitute the initial value into formula (5) to reduce the amount of calculation, use the least squares method to calculate the point cloud sub-image at this moment and the point cloud sub-image at the previous moment in pairs, and finally obtain the matching pose (x0, y0, θ0) of the lidar.

5. The method for constructing a stereoscopic parking garage map in a navigation system according to claim 4, characterized in that: In step (4), the generation method of the fused point cloud subgraph includes: Use the depthimage_to_laserscan function in the ROS package to divide the image data acquired by the camera into sub-images of equal size. The depth information of the sub-image is converted into the length from the focal point. The position of the sub-image in the point cloud sub-image is calculated by the intrinsic parameter matrix and the extrinsic parameter matrix. The calculation formula of the camera coordinate point (u, v) and the spatial coordinate point (X, Y, Z) is: where f x and f y is the focal length information, c x and c y is the main point information, R is the rotation information matrix, and t is the translation matrix.

6. The method for constructing a stereoscopic parking garage map in a navigation system according to claim 5, characterized in that: In step (5), the ground point cloud is stripped in the following manner: Set the ground height threshold to d, and place all points in the fused point cloud sub-graph whose absolute value of Z-axis height is less than d in set S. Set S is the candidate set of ground information features. Calculate the smoothness of the points in set S. After removing outliers, set S can be considered as the stripped ground information point cloud.

7. The method for constructing a stereoscopic parking garage map in a navigation system according to claim 6, characterized in that: In step (5), the ground point cloud is stripped in the following manner: The key frames of the image data acquired by the camera are extracted, and then the RGB image is converted into an HSV image. After Gaussian filtering and image enhancement, it is converted back into an RGB image to improve image differentiation. The image is then grayscaled and binarized to highlight the edge information of the parking space lines in the image. After that, connected area marking is used to strengthen the connection between the edge information. Then, edge detection is used to extract the contour information of the parking space lines. Finally, dilation and erosion processing is used to construct the complete parking space line information based on the contour.

8. The method for constructing a stereoscopic parking garage map in a navigation system according to claim 7, characterized in that: In step (6), the method of using the updated fused point cloud subgraph to supplement the navigation system map of the car includes: A linked list is set to save the point cloud subgraph and the matching pose at each moment. When the consecutive frames after a certain moment successfully complete the matching of step (3), the dependency between the previous and next frames is realized, and the pose is corrected according to the time difference between the key frames.

Images