Real-time denoising method and system for laser point cloud data

Abstract

The application discloses a kind of laser point cloud data real-time denoising method and system, and the application relates to the technical field of data denoising method, determine abnormal laser point cloud data according to the actual image of arc cover, multiple sub-abnormal regions and corresponding abnormal position, determine the real-time denoising measure of abnormal laser point cloud data based on the morphology of abnormal laser point cloud data and the surface profile of arc cover, improve the accuracy of real-time denoising measure of abnormal laser point cloud data.Therefore, based on the influencing factors of the real-time denoising process of laser point cloud data determine the influence factor of denoising efficiency;According to the influence factor, untreated abnormal laser point cloud data and the data load of laser radar, determine the optimization measure of real-time denoising, and trigger the dynamic optimization of real-time denoising process of laser point cloud data, to improve the real-time denoising efficiency of laser point cloud data.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of data denoising methods, and in particular to a laser point cloud data real-time denoising method and system. BACKGROUND

[0002] With the development of science and technology, the laser radar can scan engineering components such as arc-shaped cover bodies as radar components to collect corresponding laser point cloud data. In the prior art, laser point cloud data is collected and denoised. However, the existing denoising methods are more based on the overall thinking to quickly denoise the laser point cloud data, and ignore the further detection of sub-abnormal regions, which cannot control the real-time denoising of abnormal laser point cloud data, affecting the accuracy of the real-time denoising measures of abnormal laser point cloud data, and further cannot guarantee the dynamic optimization of the real-time denoising process of laser point cloud data. SUMMARY

[0003] The present application provides a laser point cloud data real-time denoising method and system.

[0004] The present application provides a laser point cloud data real-time denoising method, comprising:

[0005] Based on the detection of the arc-shaped cover body by the laser radar, a plurality of laser points are collected, and based on the synthesis of the plurality of laser points, laser point cloud data is determined;

[0006] According to the division of the laser point cloud data, a plurality of point cloud regions are determined, based on the region detection of each point cloud region, a plurality of sub-abnormal regions are determined, and the corresponding abnormal positions are marked;

[0007] According to the actual image of the arc-shaped cover body, the plurality of sub-abnormal regions and the corresponding abnormal positions, the abnormal laser point cloud data is determined, based on the morphology of the abnormal laser point cloud data and the surface profile of the arc-shaped cover body, the real-time denoising measures of the abnormal laser point cloud data are determined, and the laser point cloud data is optimized as a whole;

[0008] The real-time denoising process of the laser point cloud data is collected, and based on the real-time denoising process of the laser point cloud data, the influencing factors affecting the denoising efficiency are determined;

[0009] According to the influencing factors, the unprocessed abnormal laser point cloud data and the data load of the laser radar, the optimization measures of real-time denoising are determined, and the dynamic optimization of the real-time denoising process of the laser point cloud data is triggered, so as to improve the real-time denoising efficiency of the laser point cloud data.

[0010] The present application provides a laser point cloud data real-time denoising system, which is applied to the above-mentioned laser point cloud data real-time denoising method, and comprises:

[0011] The laser point cloud data module is used to collect multiple laser points based on the detection of the arc-shaped cover by the lidar, and to determine the laser point cloud data based on the synthesis of multiple laser points;

[0012] The sub-anomaly region module is used to determine multiple point cloud regions based on the division of laser point cloud data, determine multiple sub-anomaly regions based on the region detection of each point cloud region, and mark the corresponding anomaly locations.

[0013] The real-time denoising module is used to determine abnormal laser point cloud data based on the actual image of the arc-shaped cover, multiple sub-abnormal regions and corresponding abnormal locations. Based on the shape of the abnormal laser point cloud data and the surface contour of the arc-shaped cover, it determines real-time denoising measures for the abnormal laser point cloud data in order to optimize the laser point cloud data as a whole.

[0014] The influencing factors module is used to collect the real-time denoising process of laser point cloud data and determine the influencing factors that affect the denoising efficiency based on the real-time denoising process of laser point cloud data.

[0015] The optimization module is used to determine optimization measures for real-time denoising based on the influencing factors, unprocessed abnormal laser point cloud data, and the data load of the lidar, and to trigger dynamic optimization of the real-time denoising process of the laser point cloud data to improve the real-time denoising efficiency of the laser point cloud data.

[0016] Compared with the prior art, the beneficial effects of the present invention are:

[0017] In this embodiment of the invention, the method involves acquiring multiple laser points based on the detection of the arc-shaped cover by a lidar, and determining laser point cloud data based on the synthesis of these multiple laser points. Multiple point cloud regions are determined based on the division of the laser point cloud data, and multiple sub-anomaly regions are determined based on region detection of each point cloud region, marking the corresponding anomaly locations. Abnormal laser point cloud data is determined based on the actual image of the arc-shaped cover, the multiple sub-anomaly regions, and the corresponding anomaly locations. Real-time denoising measures for the abnormal laser point cloud data are determined based on the morphology of the abnormal laser point cloud data and the surface contour of the arc-shaped cover, thus performing overall optimization of the laser point cloud data. This approach incorporates abnormal laser point cloud data, taking into account both the morphology of the abnormal laser point cloud data and the overall surface contour of the arc-shaped cover, thereby improving the accuracy of the real-time denoising measures for the abnormal laser point cloud data.

[0018] Therefore, the real-time denoising process of acquired laser point cloud data is determined based on the real-time denoising process of the laser point cloud data, identifying the factors affecting denoising efficiency. Based on these factors, unprocessed abnormal laser point cloud data, and the data load of the lidar, optimization measures for real-time denoising are determined, triggering dynamic optimization of the real-time denoising process of the laser point cloud data. This further controls the real-time denoising process of the laser point cloud data, introduces factors affecting denoising efficiency, and achieves a holistic consideration of these factors, unprocessed abnormal laser point cloud data, and the data load of the lidar. This ensures dynamic optimization of the real-time denoising process of the laser point cloud data, improving its real-time denoising efficiency and ensuring the accuracy of identification and dynamic optimization effects of high-precision real-time denoising measures for laser point cloud data. Attached Figure Description

[0019] Figure 1 This is a flowchart illustrating the real-time denoising method for laser point cloud data in an embodiment of the present invention.

[0020] Figure 2 This is a flowchart illustrating step S11 in the real-time denoising method for laser point cloud data in an embodiment of the present invention.

[0021] Figure 3 This is a flowchart illustrating step S12 in the real-time denoising method for laser point cloud data in an embodiment of the present invention.

[0022] Figure 4 This is a flowchart illustrating step S13 in the real-time denoising method for laser point cloud data in an embodiment of the present invention.

[0023] Figure 5 This is a flowchart illustrating step S14 in the real-time denoising method for laser point cloud data in this embodiment of the invention.

[0024] Figure 6 This is a flowchart illustrating step S15 in the real-time denoising method for laser point cloud data in an embodiment of the present invention.

[0025] Figure 7 This is a schematic diagram of the structural composition of the real-time laser point cloud data denoising system in an embodiment of the present invention. Detailed Implementation

[0026] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention.

[0027] Please see Figures 1 to 7 A real-time denoising method for laser point cloud data is provided, applicable to denoising scenarios involving laser point cloud data. The real-time denoising method for laser point cloud data includes:

[0028] Step S11: Collect multiple laser points based on the detection of the arc-shaped cover by the lidar, and determine the laser point cloud data based on the synthesis of multiple laser points;

[0029] Step S12: Determine multiple point cloud regions based on the division of laser point cloud data, determine multiple sub-anomaly regions based on the region detection of each point cloud region, and mark the corresponding anomaly locations;

[0030] Step S13: Determine abnormal laser point cloud data based on the actual image of the arc-shaped cover, multiple sub-abnormal regions and corresponding abnormal locations. Determine real-time denoising measures for the abnormal laser point cloud data based on the shape of the abnormal laser point cloud data and the surface contour of the arc-shaped cover, so as to optimize the laser point cloud data in an overall manner.

[0031] Step S14: Collect the real-time denoising process of laser point cloud data, and determine the influencing factors affecting the denoising efficiency based on the real-time denoising process of laser point cloud data;

[0032] Step S15: Determine optimization measures for real-time denoising based on the influencing factors, unprocessed abnormal laser point cloud data, and the data load of the lidar, and trigger dynamic optimization of the real-time denoising process of the laser point cloud data to improve the real-time denoising efficiency of the laser point cloud data.

[0033] refer to Figure 2 In step S11, multiple laser points are collected based on the detection of the arc-shaped cover by the lidar, and laser point cloud data is determined based on the synthesis of multiple laser points.

[0034] In the specific implementation of this invention, the specific steps are as follows:

[0035] S111: When the lidar detects the arc-shaped cover, it collects multiple laser points corresponding to the arc-shaped cover and presents multiple laser points in the same coordinate system, with the multiple laser points distributed at different positions in the coordinate system.

[0036] S112: In the same coordinate system, multiple discrete laser points are determined based on the identification of multiple laser points. The positions corresponding to the multiple discrete laser points are the outer peripheral positions of the arc-shaped cover. Laser point cloud data is determined based on the optimization of multiple discrete laser points. This laser point cloud data does not contain multiple discrete laser points.

[0037] In the embodiments of this application, when the lidar detects the arc-shaped cover, it collects multiple laser points corresponding to the arc-shaped cover and presents multiple laser points in the same coordinate system. The multiple laser points are distributed at different positions in the coordinate system, thus introducing the distribution of multiple laser points at different positions in the coordinate system.

[0038] At this point, a lidar is used to scan the curved cover, acquiring multiple laser points on its surface. A suitable lidar is selected based on the size, shape, and inspection requirements of the curved cover. For example, if the curved cover is small and has rich surface details, a high-resolution lidar should be chosen; if the curved cover is large and the inspection environment is complex, a lidar with a wide scanning range can be selected. The lidar's scanning parameters are set, including scanning angle, scanning frequency, and resolution. For example, for a curved cover with a diameter of 1 meter, the scanning angle can be set to 360°, the scanning frequency to 100Hz, and the resolution to 0.1°. The lidar is placed in a suitable position to ensure its scanning range completely covers the curved cover. The lidar is then activated to begin scanning the curved cover.

[0039] Laser points related to the curved cover are extracted from the scanning data acquired by the lidar. During the scanning process, the lidar emits a laser beam and measures the return time (time-of-flight method) or phase difference (phase method) to calculate the three-dimensional coordinates (x, y, z) of each laser point. The acquired laser point data is stored in a point cloud format, such as PLY, LAS, or ASCII. Each point cloud contains multiple laser points, each with three coordinate values ​​(x, y, z). The acquired laser points may contain background noise (such as reflections from the surrounding environment) or invalid data points. These noisy points can be removed by setting thresholds (such as distance thresholds or intensity thresholds).

[0040] Multiple acquired laser points are unified into the same coordinate system for subsequent processing. A suitable coordinate system, such as a Cartesian coordinate system, is selected. For curved covers, a coordinate system with the center of the curved cover as the origin can be chosen. The coordinates of the laser points acquired by the lidar are then transformed into the selected coordinate system. If the lidar coordinate system differs from the target coordinate system, coordinate transformation is required. For example, the point cloud data can be transformed from the lidar coordinate system to the target coordinate system using a rotation matrix and a translation vector. The transformed laser point cloud data is then presented in 3D visualization software (such as CloudCompare or MeshLab) for intuitive observation of the laser point distribution.

[0041] Optionally, suppose we are inspecting an arc-shaped cover with a diameter of 1 meter, resembling a semi-circular lid, mounted on an industrial device; select a high-resolution LiDAR, such as the Velodyne VLP-16, with a 360° scanning range, a vertical field of view of ±15°, and a resolution of 0.1°; scan angle: 360°; scan frequency: 100Hz; resolution: 0.1°; place the LiDAR 1 meter away from the arc-shaped cover, ensuring its scanning range completely covers it. Start the LiDAR and begin scanning.

[0042] During the scanning process, the lidar emits laser beams, measures the return time, and calculates the three-dimensional coordinates (x, y, z) of each laser point. Assume that after the scanning is completed, the collected lidar point cloud data contains 10,000 points; the collected lidar point cloud data is stored in the PLY format, and each point contains three coordinate values (x, y, z); by setting a distance threshold (for example, only retaining points within 1 meter from the lidar) and an intensity threshold (removing noise points with too low reflection intensity), the laser points related to the arc cover are screened out. Assume that 8,000 valid points remain after screening.

[0043] Select a Cartesian coordinate system with the center of the arc cover as the origin; assume that there is a certain rotation and translation relationship between the coordinate system of the lidar and the target coordinate system. By calculating the rotation matrix and translation vector, the lidar point cloud data is transformed from the lidar coordinate system to the target coordinate system; the transformed lidar point cloud data is imported into the CloudCompare software to observe the distribution of the laser points. It can be seen that the laser points are evenly distributed on the surface of the arc cover, forming a semi-circular point cloud.

[0044] Furthermore, in the same coordinate system, multiple discrete laser points are determined based on the recognition of multiple laser points. The positions corresponding to the multiple discrete laser points are the outer peripheral positions of the arc cover. Based on the optimization of the multiple discrete laser points by multiple laser points, the lidar point cloud data is determined. This lidar point cloud data does not contain multiple discrete laser points, taking into account the overall optimization of the multiple discrete laser points by multiple laser points, ensuring the accuracy of the lidar point cloud data.

[0045] At this time, discrete laser points are identified in the lidar point cloud data. These points are usually located at the outer peripheral positions of the arc cover. The discrete laser points usually have a relatively large local density difference from other points. By calculating the distance between each point and its neighboring points, the points with low density can be identified; select a suitable neighborhood radius, for example, neighborhood radius = 0.05 meters (adjusted according to the size of the arc cover and the resolution of the lidar); for each laser point P i , calculate its distance from other points in the neighborhood and count the number of points N in the neighborhood i ; set a density threshold T, for example, T = 5 (indicating that there should be at least 5 points in the neighborhood of each point). If N i < T, then P i is considered a discrete laser point.

[0046] Discrete laser points usually have a relatively high curvature or a relatively large difference in normal vector from other points. Calculate the normal vector and curvature of each laser point. For the arc cover, the points with higher curvature may be discrete laser points; set a curvature threshold C, for example, C = 0.5 (adjusted according to the curvature range of the arc cover). If the curvature C of a certain pointi If >C, then the point is considered to be a discrete laser point.

[0047] The identified discrete laser points were optimized to determine the final laser point cloud data, ensuring that this data did not contain discrete laser points. This involved simple removal: directly deleting the identified discrete laser points; smoothing: replacing discrete laser points with the average or median of their neighboring points; and interpolation methods: if there were many discrete laser points, interpolation methods (such as linear interpolation, spline interpolation, etc.) could be used to fill in their positions. This successfully identified and optimized the discrete laser points in the laser point cloud data of the curved cover. The final laser point cloud data does not contain discrete laser points and can more accurately reflect the geometric features of the curved cover.

[0048] refer to Figure 3 In step S12, multiple point cloud regions are determined based on the division of laser point cloud data, multiple sub-abnormal regions are determined based on the region detection of each point cloud region, and the corresponding abnormal locations are marked.

[0049] In the specific implementation of this invention, the specific steps are as follows:

[0050] S121: In laser point cloud data, multiple point cloud concentration areas are determined based on the division of laser point cloud data, and multiple point cloud regions are determined based on the location of multiple point cloud concentration areas and the connection between two adjacent point cloud concentration areas.

[0051] S122: Simultaneously detect multiple point cloud regions. At this time, multiple abnormal point cloud features are determined based on the region detection of each point cloud region.

[0052] S123: Based on the location of multiple abnormal point cloud features and the regional distribution map of the point cloud region, determine multiple sub-abnormal regions, and determine the corresponding abnormal location based on the location detection of multiple sub-abnormal regions.

[0053] In the embodiments of this application, multiple point cloud concentration areas are determined based on the division of the laser point cloud data. Multiple point cloud regions are determined based on the positions of the multiple point cloud concentration areas and the connection between two adjacent point cloud concentration areas. This approach takes into account both the positions of the multiple point cloud concentration areas and the connection between two adjacent point cloud concentration areas, thus ensuring the accuracy of the multiple point cloud regions.

[0054] At this point, the laser point cloud data is divided into multiple point cloud clusters, and several partitioning methods are introduced, including voxel grids, octrees, and density-based clustering. Voxel grids divide the space into fixed-size cubes (voxels), with points within each voxel considered as a point cloud cluster. This method is simple and efficient, suitable for large-scale point cloud data. Octrees recursively divide the space into eight sub-regions, with points within each sub-region forming a point cloud cluster. This method is suitable for handling sparse data, but has high computational complexity. Density-based clustering divides the point cloud data into multiple regions based on point density. This method can automatically identify regions with different densities, but requires parameter adjustment (such as neighborhood radius and minimum number of points).

[0055] Choose an appropriate voxel size. For example, for an arc-shaped cover with a diameter of 1 meter, a voxel size of 0.05 meters can be selected; for density-based clustering methods, set the neighborhood radius (e.g., 0.05 meters) and the minimum number of points (e.g., 5); map the point cloud data into a voxel grid, with points within each voxel forming a concentrated region of the point cloud; determine multiple point cloud regions based on the location of the concentrated region of the point cloud and the connectivity of adjacent regions.

[0056] If the distance between two voxels is less than a certain threshold (e.g., twice the voxel size), the two voxels are considered connected, and all connected voxels are merged into a single point cloud region. Optionally, assuming we have completed step S11 and obtained laser point cloud data containing 8,000 points, we select a voxel mesh generation method and choose a voxel size of 0.05 meters; we then map the 8,000 points into the voxel mesh. Assume that after generation, we obtain 1,000 voxels (the concentrated point cloud region).

[0057] Set the connection threshold to twice the voxel size, i.e., 0.1 meters. Check the distance between each voxel and other voxels; if the distance is less than 0.1 meters, they are considered connected. Merge all connected voxels into a single point cloud region. Assuming a final result of 100 point cloud regions, collect a point cloud region matching table, as shown in Table 1:

[0058] Table 1: Cloud Concentration Area Matching Table

[0059]

[0060] Furthermore, multiple point cloud regions are detected simultaneously. At this point, multiple abnormal point cloud features are determined based on the region detection of each point cloud region, which takes into account the overall consideration of region detection of each point cloud region and ensures the accuracy of multiple abnormal point cloud features.

[0061] At this point, multiple point cloud regions can be detected simultaneously to improve efficiency. Multithreading or multiprocessing techniques can be used to process multiple point cloud regions concurrently. For example, point cloud regions can be assigned to different threads or processes, with each thread or process handling one region independently. This can be achieved using Python's `multiprocessing` module or C++'s multithreading library (such as `std::thread`). Alternatively, the point cloud data can be divided into multiple subsets, each corresponding to a point cloud region. These subsets can then be assigned to different processing units (threads or processes). Optionally, if there are 100 point cloud regions, 10 threads can be started, with each thread processing 10 regions.

[0062] Anomaly features are identified in each point cloud region. The local density of each point is calculated, for example, by counting the number of points within each point and its neighbors. The curvature of each point is calculated; points with high curvature are likely anomalies. The normal vector of each point is calculated, and its consistency with the normal vectors of its neighbors is checked. Points with significant differences in normal vectors are likely anomalies. A density threshold is set, for example, Dthreshold = 0.2. If the local density of a point is below this threshold, it is considered an anomaly. A curvature threshold is set, for example, Cthreshold = 0.5. If the curvature of a point is above this threshold, it is considered an anomaly. A normal vector consistency threshold is set, for example, θthreshold = 30°. If the angle between the normal vector of a point and the normal vectors of its neighbors is greater than this threshold, it is considered an anomaly. Multiple features (density, curvature, and normal vector consistency) are combined for a comprehensive judgment. Weights can be assigned to each feature, and a comprehensive score can be calculated.

[0063] Si = w1 × Di + w2 × Ci + w3 × θi

[0064] Where w1, w2, w3 are weights, and Di, Ci, θi are the density score, curvature score, and normal vector consistency score of point i, respectively.

[0065] Therefore, multiple sub-anomaly regions are determined based on the location of multiple anomalous point cloud features and the regional distribution map of the point cloud regions. The corresponding anomaly location is determined based on the location detection of multiple sub-anomaly regions, which takes into account the overall consideration of the location detection of multiple sub-anomaly regions and ensures the accuracy of the corresponding anomaly location.

[0066] At this point, based on the location of the abnormal point cloud features, the area where abnormal points are clustered is divided into sub-abnormal regions. Clustering algorithms (such as DBSCAN, K-Means, etc.) are used to cluster the abnormal points, and adjacent abnormal points are grouped into a sub-abnormal region. For the DBSCAN algorithm, the neighborhood radius (e.g., 0.05 meters) and the minimum number of points (e.g., 5) are set. Based on the clustering results, each cluster is considered as a sub-abnormal region, and each sub-abnormal region contains multiple abnormal points.

[0067] Determine the center location or other key locations of each sub-anomaly region as the anomaly location, and calculate the average coordinates of all points within each sub-anomaly region to determine the center location of that sub-anomaly region.

[0068]

[0069] Where (xi,yi,zi) are the coordinates of points within the sub-anomaly region, and n is the number of points;

[0070] On the distribution map of the point cloud region, mark the center position of each sub-anomaly region to obtain multiple sub-anomaly regions and their corresponding anomaly positions. Optionally, use the DBSCAN algorithm to cluster the anomaly points, with a neighborhood radius of ∈ = 0.05 meters and a minimum number of points minPts = 5. Clustering results: Sub-anomaly region 1 contains points 1 and 3 from point cloud region 1; Sub-anomaly region 2 contains points 1 and 2 from point cloud region 2; Sub-anomaly region 3 contains points 1 and 3 from point cloud region 3. Collect a sub-anomaly region matching table, which is shown in Table 2.

[0071] Table 2 Sub-anomaly Region Matching Table

[0072]

[0073] refer to Figure 4 In step S13, abnormal laser point cloud data is determined based on the actual image of the arc-shaped cover, multiple sub-abnormal regions and corresponding abnormal positions. Real-time denoising measures for the abnormal laser point cloud data are determined based on the shape of the abnormal laser point cloud data and the surface contour of the arc-shaped cover, so as to optimize the laser point cloud data in an overall manner.

[0074] In the specific implementation of this invention, the specific steps are as follows:

[0075] S131: Monitor the arc-shaped cover in real time and collect actual images of the arc-shaped cover; determine multiple first-anomaly point cloud data based on the actual images of the arc-shaped cover and multiple sub-anomaly regions;

[0076] S132: Determine multiple second abnormal point cloud data based on the actual image of the arc-shaped cover and the corresponding abnormal location, and determine abnormal laser point cloud data based on the matching of multiple first abnormal point cloud data and multiple second abnormal point cloud data;

[0077] S133: Collect the shape of abnormal laser point cloud data, determine the corresponding point cloud anomaly type by comparing the shape of the abnormal laser point cloud data with the surface contour of the arc-shaped cover, determine the real-time denoising measures for the abnormal laser point cloud data based on the mapping relationship between the point cloud anomaly type and the real-time denoising measures, and trigger the overall optimization of the laser point cloud data based on the execution of the real-time denoising measures for the abnormal laser point cloud data.

[0078] In the embodiments of this application, the curved cover is monitored in real time and the actual image of the curved cover is acquired; multiple first abnormal point cloud data are determined based on the actual image of the curved cover and multiple sub-abnormal regions, which takes into account the overall consideration of the actual image of the curved cover and multiple sub-abnormal regions, and ensures the accuracy of the multiple first abnormal point cloud data.

[0079] At this point, the curved cover is monitored in real time via a camera, and its actual images are captured. A suitable camera is selected, such as a high-definition industrial camera or a regular RGB camera, ensuring that the camera's resolution, frame rate, and field of view can cover the entire curved cover. A high-definition camera with a resolution of 1920×1080 and a frame rate of 30fps is chosen. The camera is installed in a position that can clearly capture the curved cover, ensuring that the curved cover is completely within the camera's field of view. The camera is calibrated to ensure the accuracy of the image data. A calibration board can be used for calibration. The camera is then started, and image data of the curved cover is captured in real time, assuming the camera captures 30 frames per second.

[0080] Combining the actual image and sub-anomaly regions, the first anomalous point cloud data is determined. The acquired image undergoes preprocessing, including grayscale conversion, binarization, and filtering, to highlight the anomalous regions. After grayscale conversion, Gaussian filtering is applied to remove noise. Image segmentation algorithms (such as thresholding, edge detection, and region growing) are used to identify anomalous regions in the image. Thresholding algorithms identify regions with brightness exceeding a certain threshold as anomalous regions. The anomalous regions in the image are then matched with the sub-anomaly regions in the point cloud data. Image coordinates can be mapped to the point cloud coordinate system through coordinate transformation. Optionally, assuming the image segmentation algorithm identifies two anomalous regions, corresponding to sub-anomaly region 1 and sub-anomaly region 2 in the point cloud, the corresponding point cloud data is extracted as the first anomalous point cloud data. In this case, the point cloud data of sub-anomaly region 1 and sub-anomaly region 2 are extracted from the point cloud data as the first anomalous point cloud data. By monitoring the curved cover in real time, we acquired its actual images and identified the first anomalous point cloud data by combining it with sub-anomaly regions. Through image preprocessing, segmentation, and fusion with the point cloud data, we were able to accurately identify the anomalous region and extract the corresponding point cloud data. Furthermore, we can adjust the camera parameters, image preprocessing methods, and segmentation algorithms according to the specific point cloud data and requirements. For example, for point cloud data of different shapes and densities, the threshold and method of the image segmentation algorithm can be adjusted to achieve the best results.

[0081] Furthermore, multiple second abnormal point cloud data are determined based on the actual image of the arc-shaped cover and the corresponding abnormal location. Abnormal laser point cloud data is determined based on the matching of multiple first abnormal point cloud data and multiple second abnormal point cloud data. This overall consideration of matching multiple first abnormal point cloud data and multiple second abnormal point cloud data ensures the accuracy of abnormal laser point cloud data.

[0082] At this point, combined with the abnormal locations in the actual image, the corresponding point cloud data is extracted as the second abnormal point cloud data. In the actual image, the abnormal regions are identified by the image segmentation algorithm, and the locations of these abnormal regions are determined. At this point, it is assumed that the image segmentation algorithm identifies two abnormal regions, located in the upper left corner and the lower right corner of the image, respectively.

[0083] Converting image coordinates to point cloud coordinates typically requires a mapping from image space to point cloud space, which can be obtained through calibration. Let's assume that calibration has provided this mapping. For example, the anomaly region in the upper left corner of the image corresponds to sub-anomaly region 1 in the point cloud, and the anomaly region in the lower right corner corresponds to sub-anomaly region 3 in the point cloud.

[0084] Based on the abnormal locations in the image, the corresponding regions in the point cloud data are extracted as the second abnormal point cloud data. At this point, the point cloud data of sub-abnormal regions 1 and 3 are extracted from the point cloud data as the second abnormal point cloud data. Further, by matching the first and second abnormal point cloud data, the final abnormal laser point cloud data is determined. Simultaneously, a suitable matching algorithm, such as the Iterative Nearest Point (ICP) algorithm or nearest neighbor matching, is selected to match the first and second abnormal point cloud data. Each point in the first abnormal point cloud data is matched with a point in the second abnormal point cloud data to find the best matching point pair. Optionally, it is assumed that the first abnormal point cloud data contains sub-abnormal regions 1 and 2, and the second abnormal point cloud data contains sub-abnormal regions 1 and 3. After matching using the ICP algorithm, it is found that sub-abnormal region 1 has matches in both sets of data, while sub-abnormal regions 2 and 3 do not have matches.

[0085] Based on the matching results, the final abnormal laser point cloud data is determined. Only abnormal regions that exist in both the first and second abnormal point cloud data are considered to be true abnormal laser point cloud data. Optionally, sub-abnormal region 1 is finally determined as abnormal laser point cloud data. Based on the actual image of the arc-shaped cover and the corresponding abnormal position, the second abnormal point cloud data is determined. By matching the first and second abnormal point cloud data, the final abnormal laser point cloud data is determined. Through the fusion and matching of image and point cloud data, we can more accurately identify abnormal regions.

[0086] Therefore, the morphology of abnormal laser point cloud data is collected, and the corresponding point cloud anomaly type is determined by comparing the morphology of the abnormal laser point cloud data with the surface contour of the curved cover. Real-time denoising measures for the abnormal laser point cloud data are then determined based on the mapping relationship between the point cloud anomaly type and real-time denoising measures. The execution of these real-time denoising measures triggers a holistic optimization of the laser point cloud data, incorporating a comprehensive consideration of the mapping relationship between point cloud anomaly types and real-time denoising measures. This ensures the accuracy of the real-time denoising measures for abnormal laser point cloud data. Furthermore, the inclusion of abnormal laser point cloud data, combined with a comprehensive consideration of the morphology of the abnormal laser point cloud data and the surface contour of the curved cover, improves the accuracy of the real-time denoising measures for abnormal laser point cloud data.

[0087] At this point, we analyze the geometric characteristics of the anomalous laser point cloud data, such as curvature, normal vector, and local density. For curvature: we calculate the curvature of each anomalous point; points with high curvature may be surface noise or edge noise. For normal vector: we calculate the normal vector of each anomalous point; points with poor normal vector consistency may be isolated noise. For local density: we calculate the local density of each anomalous point; points with low density may be isolated noise. Based on the calculated characteristics, we analyze the overall shape of the anomalous point cloud data, assuming that the anomalous laser point cloud data has high curvature, poor normal vector consistency, and low local density.

[0088] By comparing the morphology of the abnormal point cloud data with the surface contour of the arc-shaped cover, the anomaly type is determined. The surface contour is extracted from the normal point cloud data of the arc-shaped cover, and its geometric features (such as curvature, normal vector, etc.) are calculated. Optionally, it is assumed that the surface contour curvature of the arc-shaped cover is low and the normal vector consistency is high.

[0089] The features of the anomalous point cloud data are compared with the features of the surface contour to determine the anomaly type. If the curvature of the anomalous point is significantly higher than that of the surface contour, it may be surface noise. If the normal vector of the anomalous point differs greatly from the normal vector of the surface contour, it may be edge noise. If the normal vector of the anomalous point differs greatly from the normal vector of the surface contour, it may be edge noise. Assuming that the curvature of the anomalous point is high, the consistency of the normal vector is poor, and the local density is low, it is finally determined to be surface noise.

[0090] Based on the anomaly type, select the appropriate real-time denoising measures and establish a mapping relationship between the anomaly type and the real-time denoising measures. Optionally, for surface noise, use curvature-based smoothing filtering; for edge noise, use normal vector-based smoothing filtering; for isolated noise, use density-based filtering (such as voxel filtering). At the same time, based on the determined anomaly type, select the appropriate denoising measures. In this case, for surface noise, select curvature-based smoothing filtering.

[0091] The selected denoising measures are applied to the abnormal laser point cloud data, and the entire laser point cloud data is optimized. The selected denoising measures are applied to the abnormal point cloud data, and the denoised abnormal point cloud data is reintegrated into the original point cloud data. The entire laser point cloud data is optimized. Optionally, curvature-based smoothing filtering is applied to the surface noise, and the corresponding part in the original data is replaced by the denoised abnormal point cloud data to obtain the optimized laser point cloud data.

[0092] Optionally, collect abnormal laser point cloud data tables, as shown in Table 3:

[0093] Table 3 Abnormal Laser Point Cloud Data Table

[0094]

[0095] Based on the above table, the curvature of the outliers is calculated, assuming a curvature of 0.6; the normal vectors of the outliers are calculated, assuming a normal vector uniformity of 30°; the local density of the outliers is calculated, assuming a local density of 0.2. At this point, the outliers have high curvature, poor normal vector uniformity, and low local density. For the arc-shaped cover surface contour: the curvature is low (e.g., 0.1), the normal vector uniformity is high (e.g., 5°), the curvature of the outliers is significantly higher than the surface contour, the normal vector uniformity is poor, and the local density is low. The outlier type is surface noise. Simultaneously, the surface noise denoising measure is curvature-based smoothing filtering. Therefore, curvature-based smoothing filtering is performed on the outlier point cloud data, and the denoised outlier point cloud data is reintegrated into the original point cloud data. Optionally, assuming that after smoothing filtering, the coordinates of the outliers are adjusted to (0.12, 0.22, 0.32) and (0.16, 0.26, 0.36), the optimized laser point cloud data table is shown in Table 4.

[0096] Table 4: Optimized Laser Point Cloud Data

[0097] Point cloud region number Point number Coordinate (x, y, z) 1 1 (0.12,0.22,0.32) 1 2 (0.16,0.26,0.36) 2 1 (0.2,0.3,0.4) 2 2 (0.25,0.3

[0098] refer to Figure 5 In step S14, the real-time denoising process of laser point cloud data is collected, and the influencing factors affecting the denoising efficiency are determined based on the real-time denoising process of laser point cloud data.

[0099] In the specific implementation of this invention, the specific steps are as follows:

[0100] S141: Monitor the execution of real-time denoising measures for abnormal laser point cloud data in real time, mark the processed abnormal laser point cloud data, and determine the real-time denoising process of laser point cloud data based on the detection of processed abnormal laser point cloud data.

[0101] S142: In the real-time denoising process of laser point cloud data, the denoising items to be processed and the denoising items being processed are determined based on the analysis of the real-time denoising process of laser point cloud data.

[0102] S143: Determine the corresponding denoising efficiency dwell range based on the detection of the denoising item being processed, determine the first sub-influencing factor based on the identification of the denoising efficiency dwell range, determine the corresponding second sub-influencing factor based on the denoising item to be processed, and determine the influencing factor affecting the denoising efficiency based on the combination of the first sub-influencing factor and the second sub-influencing factor.

[0103] In the embodiments of this application, the execution of real-time denoising measures for abnormal laser point cloud data is monitored in real time, and the processed abnormal laser point cloud data is marked. The real-time denoising process of the laser point cloud data is determined based on the detection of the processed abnormal laser point cloud data, which is compatible with the overall consideration of the detection of processed abnormal laser point cloud data and ensures the accuracy of the real-time denoising process of the laser point cloud data.

[0104] At this point, monitor the execution of denoising measures in real time to ensure that the denoising process proceeds as planned. Select appropriate monitoring tools, such as point cloud processing software (e.g., PCL, CloudCompare) or custom monitoring scripts. Optionally, use the real-time monitoring function in the PCL library.

[0105] Set monitoring metrics, such as processing time, processing progress, and processing status, to monitor the processing time of each abnormal point cloud data and record the processing progress. Collect execution data of denoising measures in real time, including processing time and processing status. At this time, data is collected once per second to record the processing status of each abnormal point cloud data.

[0106] Distinguish between processed and unprocessed anomalous point cloud data, add a label such as "processed" to the processed anomalous point cloud data, and automatically update the label after the denoising measures are completed. At the same time, based on the detection results of the processed data, determine the real-time denoising progress of the entire laser point cloud data, detect the processed anomalous point cloud data, and count the proportion of processed data. Based on the statistical results, determine the status of the real-time denoising process. Optionally, if the processed data accounts for 50% of the total anomalous data, the denoising process is considered to be half completed.

[0107] Furthermore, in the real-time denoising process of laser point cloud data, the denoising items to be processed and the denoising items in process are determined based on the analysis of the real-time denoising process of laser point cloud data. This takes into account the overall consideration of the analysis of the real-time denoising process of laser point cloud data and ensures the accuracy of the denoising items to be processed and the denoising items in process.

[0108] At this point, the data stream in the real-time denoising process is analyzed to identify the current processing status. The real-time monitoring function of the PCL library is used to record the processing status and processing time of each abnormal point cloud data. The monitoring log is parsed to extract the processing status (pending processing, processing, processed) of each abnormal point cloud data. At this point, the log file is parsed to extract the processing status and processing time of each abnormal point cloud data.

[0109] Identify anomalous point cloud data that has not yet been processed, filter out anomalous point cloud data marked "to be processed" from the parsed data stream, and generate a list of denoising projects to be processed, including the anomalous point cloud data number and the corresponding point cloud region number. At the same time, identify anomalous point cloud data that is currently being processed, filter out anomalous point cloud data marked "in processing" from the parsed data stream, and generate a list of denoising projects in processing, including the anomalous point cloud data number and the corresponding point cloud region number.

[0110] Therefore, the corresponding denoising efficiency range is determined based on the detection of the denoising items being processed. The first sub-influencing factor is determined based on the identification of the denoising efficiency range. The corresponding second sub-influencing factor is determined based on the denoising items to be processed. The influencing factors affecting denoising efficiency are determined based on the combination of the first and second sub-influencing factors. This method takes into account the overall consideration of the combination of the first and second sub-influencing factors, ensuring the accuracy of the influencing factors affecting denoising efficiency.

[0111] At this point, analyze the denoising projects being processed, determine the processing time interval for each project, i.e. the denoising efficiency dwell interval, and record the start time and current processing time of each denoising project being processed. Optionally, for abnormal point cloud data number 2, record its start time as 10:00:01, current time as 10:00:02, and processing time as 1 second.

[0112] Based on the processing time, the denoising efficiency dwell range for each item is determined. The processing time can be divided into different intervals, such as 0-1 seconds, 1-2 seconds, 2-3 seconds, etc. In this case, the processing time for abnormal point cloud data number 2 is 1 second, falling within the 0-1 second dwell range. Simultaneously, the denoising efficiency dwell range is analyzed to determine the main factors affecting denoising efficiency (the first sub-influencing factor). The number of items within each dwell range is counted, and the intervals with longer processing times are analyzed. Based on the statistical results, the main factors affecting denoising efficiency are identified. For example, items with longer processing times may involve high-density noise or complex geometric features. Optionally, statistics show that there is 1 item in the 0-1 second interval, 0 items in the 1-2 second interval, and 0 items in the 2-3 second interval. This indicates that items with longer processing times typically involve high-density noise; therefore, the first sub-influencing factor is "high-density noise." At this point, the first sub-influencing factor is the key bottleneck factor identified based on the real-time performance data (such as the processing time interval) of the denoising project being processed, reflecting the core problem that is slowing down the denoising efficiency in the current execution; the second sub-influencing factor is the potential efficiency obstacle predicted based on the data characteristics (such as noise type and data volume) of the denoising project to be processed, reflecting the expected impact of subsequent tasks on the denoising efficiency.

[0113] Analyze the denoising project to be processed to identify factors that may affect the denoising efficiency (second sub-influencing factor). Analyze the characteristics of the project to be processed, such as data volume, noise type, geometric complexity, etc. Based on the characteristics of the project to be processed, identify factors that may affect the denoising efficiency. Optionally, the abnormal point cloud data number 3 to be processed has a large data volume, containing 1000 points. It is found that point clouds with a large data volume in the project to be processed require a longer processing time. Therefore, the second sub-influencing factor is "large data volume".

[0114] By combining the first and second sub-influencing factors, the comprehensive factors affecting denoising efficiency are determined. A comprehensive analysis of the first and second sub-influencing factors is conducted to identify the comprehensive influencing factor. Based on the comprehensive analysis results, the comprehensive factors affecting denoising efficiency are determined. Optionally, the first sub-influencing factor is "high-density noise," the second sub-influencing factor is "large data volume," and the comprehensive influencing factor is "high-density noise and large data volume." The analysis of denoising projects under processing and those yet to be processed identifies the first and second sub-influencing factors affecting denoising efficiency, and the comprehensive factors affecting denoising efficiency are determined by combining these factors. This method helps us to more comprehensively understand the influencing factors of denoising efficiency, thereby optimizing denoising strategies.

[0115] refer to Figure 6 In step S15, optimization measures for real-time denoising are determined based on the influencing factors, unprocessed abnormal laser point cloud data, and the data load of the lidar, and dynamic optimization of the real-time denoising process of the laser point cloud data is triggered to improve the real-time denoising efficiency of the laser point cloud data.

[0116] In the specific implementation of this invention, the specific steps are as follows:

[0117] S151: Collect various influencing factors, determine the first optimization coefficient based on the influencing factors and unprocessed abnormal laser point cloud data, determine the second optimization coefficient based on the influencing factors and the data load of the lidar, and determine the real-time denoising optimization measures based on the mapping relationship between the first optimization coefficient, the second optimization coefficient, and the optimization measures.

[0118] S152: Real-time denoising process for acquiring laser point cloud data, and determining the node to be optimized in the real-time denoising process based on the real-time denoising process and the unprocessed abnormal laser point cloud data. The node to be optimized is matched with the optimization measures for real-time denoising.

[0119] S153: When the node to be optimized in the real-time denoising process is responded to, the optimization measures of real-time denoising are triggered, and the real-time denoising process of laser point cloud data is dynamically optimized to gradually optimize the corresponding influencing factors and improve the real-time denoising efficiency of laser point cloud data.

[0120] In the embodiments of this application, various influencing factors are collected, and a first optimization coefficient is determined based on the influencing factors and unprocessed abnormal laser point cloud data. A second optimization coefficient is determined based on the influencing factors and the data load of the lidar. Real-time denoising optimization measures are determined based on the mapping relationship between the first optimization coefficient, the second optimization coefficient, and the optimization measures. This approach takes into account the overall consideration of the mapping relationship between the first optimization coefficient, the second optimization coefficient, and the optimization measures, ensuring the accuracy of the real-time denoising optimization measures.

[0121] At this point, factors affecting denoising efficiency are collected, which may include noise type, data volume, hardware performance, etc. The processed and unprocessed abnormal point cloud data are analyzed to identify the main noise types, such as high-density noise and isolated noise. The data volume of the unprocessed abnormal point cloud data is assessed, including the number of points and data complexity. The current data load of the LiDAR is assessed, including processing speed and memory usage. Optionally, through analysis, it is found that the main noise type is high-density noise, the unprocessed point cloud data number 3 contains 1000 points, the data volume is large, the current data load of the LiDAR is high, and the processing speed is slow.

[0122] Based on the influencing factors and the unprocessed abnormal laser point cloud data, a first optimization coefficient is calculated. A weight is assigned to each influencing factor, reflecting its impact on denoising efficiency. The first optimization coefficient is calculated based on the weights of the influencing factors and the characteristics of the unprocessed data. Optionally, the weight for high-density noise is 0.6, and the weight for large data volume is 0.4. For unprocessed abnormal point cloud data number 3, the impact of high-density noise is 0.8, and the impact of large data volume is 0.7. Therefore, the first optimization coefficient = 0.6 × 0.8 + 0.4 × 0.7 = 0.48 + 0.28 = 0.76. Optionally, the first optimization coefficient is a comprehensive index reflecting the impact of the characteristics of the unprocessed abnormal laser point cloud data (such as noise type and data volume) on denoising efficiency, used to quantify the optimization needs of the current data itself.

[0123] Furthermore, a real-time denoising process for acquiring laser point cloud data is implemented. Based on the real-time denoising process and unprocessed abnormal laser point cloud data, nodes to be optimized in the real-time denoising process are determined. These nodes to be optimized are matched with the optimization measures for real-time denoising, taking into account both the real-time denoising process and the unprocessed abnormal laser point cloud data as a whole, thus ensuring the accuracy of the nodes to be optimized in the real-time denoising process.

[0124] At this point, based on the influencing factors and the data load of the LiDAR, a second optimization coefficient is calculated. A weight is assigned to each influencing factor, reflecting its impact on denoising efficiency. The second optimization coefficient is calculated based on the weights of the influencing factors and the data load of the LiDAR. Optionally, high-density noise has a weight of 0.6, and large data volume has a weight of 0.4. Given the high data load of the LiDAR, the impact of high-density noise is 0.7, and the impact of large data volume is 0.9. Therefore, the second optimization coefficient = 0.6 × 0.7 + 0.4 × 0.9 = 0.42 + 0.36 = 0.78. The second optimization coefficient reflects the optimization requirements resulting from the combined effects of the LiDAR hardware load (such as processing speed and memory usage) and data characteristics, emphasizing the adaptability of the system's real-time performance.

[0125] Based on the mapping relationship between optimization coefficients and optimization measures, optimization measures for real-time denoising are selected. A mapping relationship between optimization coefficients and optimization measures is established. Based on the first and second optimization coefficients, corresponding optimization measures are selected. Optionally, in the mapping relationship between optimization coefficients and optimization measures, for high-density noise, curvature-based smoothing filtering is used; for large amounts of data, multi-threaded processing is used. The first optimization coefficient is 0.76, and the second optimization coefficient is 0.78. The optimization measures selected are curvature-based smoothing filtering and multi-threaded processing. Simultaneously, factors affecting denoising efficiency were successfully collected. Based on these factors and relevant data, the first and second optimization coefficients were determined, and corresponding optimization measures were selected. This method helps us to select optimization measures more scientifically and improve denoising efficiency.

[0126] Therefore, when the node to be optimized in the real-time denoising process is responded to, the execution of real-time denoising optimization measures is triggered, and the real-time denoising process of laser point cloud data is dynamically optimized to gradually optimize the corresponding influencing factors and improve the real-time denoising efficiency of laser point cloud data. The real-time denoising efficiency of laser point cloud data is introduced, and the influencing factors affecting the denoising efficiency are also introduced. The overall consideration of these influencing factors, unprocessed abnormal laser point cloud data, and the data load of the lidar is realized, ensuring the dynamic optimization of the real-time denoising process of laser point cloud data to improve the real-time denoising efficiency of laser point cloud data. It is compatible with the recognition accuracy and dynamic optimization effect of real-time denoising measures for high-laser point cloud data.

[0127] At this point, monitor the denoising process in real time, record the processing status and processing time of each abnormal point cloud data, select appropriate monitoring tools, such as point cloud processing software (e.g., PCL, CloudCompare) or custom monitoring scripts, set monitoring indicators, such as processing time, processing progress, processing status, etc., and collect data in the denoising process in real time, including processing time, processing status, etc. Optionally, use the real-time monitoring function of the PCL library to monitor the processing time of each abnormal point cloud data and record the processing progress, collect data once per second, and record the processing status of each abnormal point cloud data.

[0128] Nodes with long processing times or low processing efficiency are identified and need optimization. The processing time of each node in the real-time denoising process is analyzed to identify nodes with long processing times. Unprocessed abnormal point cloud data is evaluated to determine its impact on the denoising process. Combining the evaluation of processing time and unprocessed data, nodes to be optimized are identified. Optionally, it is found that the processing time of abnormal point cloud data No. 3 is significantly longer. Unprocessed abnormal point cloud data No. 3 contains 1000 points, which is a large amount of data and may affect the overall denoising efficiency. Therefore, abnormal point cloud data No. 3 is identified as a node to be optimized.

[0129] To ensure the identified nodes to be optimized match the determined optimization measures, review the optimization measures determined in step S151 to ensure their applicability to the nodes to be optimized, verify the matching degree between the nodes to be optimized and the optimization measures, and ensure that the optimization measures can effectively solve the node problems. Optionally, review the optimization measures to confirm that curvature-based smoothing filtering and multi-threaded processing are applicable to abnormal point cloud data number 3, and verify that the high-density noise and large data volume problems of abnormal point cloud data number 3 can be solved by curvature-based smoothing filtering and multi-threaded processing. Simultaneously, the real-time denoising process of the laser point cloud data was collected, and the nodes requiring optimization (nodes to be optimized) were identified. By analyzing the processing time and evaluating the unprocessed data, we determined the nodes to be optimized and ensured that these nodes matched the determined optimization measures. This method can help us optimize the denoising process more effectively and improve denoising efficiency. Furthermore, the selection of monitoring tools, monitoring indicators, and optimization measures can be adjusted according to the specific point cloud data and requirements. For example, for different types of noise or data volume, the matching method of monitoring indicators and optimization measures can be further refined to optimize the denoising process more accurately.

[0130] Optionally, if abnormal point cloud data number 3 is detected as a node to be optimized, its processing time is expected to be long, and the data volume is 1000 points, the monitoring system is set to automatically call the optimization script or function when the node to be optimized is detected. Curvature-based smoothing filtering and multi-threaded processing are selected as optimization measures. Curvature-based smoothing filtering and multi-threaded processing are performed on abnormal point cloud data number 3. When performing curvature-based smoothing filtering, the filtering intensity is adjusted according to real-time feedback. When performing multi-threaded processing, the number of threads is dynamically adjusted according to the system load. The processing time of abnormal point cloud data number 3 is evaluated in real time by monitoring tools. It is reduced from 3 seconds to 1.5 seconds, and the optimization effect is significant. Through multiple iterations of optimization, the noise density of abnormal point cloud data number 3 is gradually reduced, and the processing time is reduced.

[0131] Meanwhile, the monitoring system detected abnormal point cloud data number 3 as a node to be optimized, automatically triggering an optimization script to perform curvature-based smoothing filtering on abnormal point cloud data number 3, adjusting the filter strength to 0.5, and starting multi-threaded processing. The number of threads was adjusted to 4 according to the system load. After the first optimization, the processing time was reduced from 3 seconds to 2 seconds. The filter strength was adjusted to 0.6, and the number of threads was adjusted to 6. After the second optimization, the processing time was reduced from 2 seconds to 1.5 seconds. The noise density was gradually optimized, and the final processing time stabilized at 1.5 seconds.

[0132] Please see Figure 7 , Figure 7 This is a schematic diagram of the structural composition of the real-time laser point cloud data denoising system in this embodiment of the invention; the real-time laser point cloud data denoising system includes:

[0133] The laser point cloud data module 21 is used to collect multiple laser points based on the detection of the arc-shaped cover by the lidar, and to determine the laser point cloud data based on the synthesis of multiple laser points;

[0134] The sub-anomaly region module 22 is used to determine multiple point cloud regions based on the division of laser point cloud data, determine multiple sub-anomaly regions based on the region detection of each point cloud region, and mark the corresponding anomaly locations.

[0135] The real-time denoising module 23 is used to determine abnormal laser point cloud data based on the actual image of the arc-shaped cover, multiple sub-abnormal regions and corresponding abnormal positions, and to determine real-time denoising measures for the abnormal laser point cloud data based on the shape of the abnormal laser point cloud data and the surface contour of the arc-shaped cover, so as to optimize the laser point cloud data in an overall manner.

[0136] Influencing Factors Module 24 is used to collect the real-time denoising process of laser point cloud data and determine the influencing factors affecting the denoising efficiency based on the real-time denoising process of laser point cloud data.

[0137] The optimization module 25 is used to determine the optimization measures for real-time denoising based on the influencing factors, unprocessed abnormal laser point cloud data and the data load of the lidar, and to trigger the dynamic optimization of the real-time denoising process of the laser point cloud data in order to improve the real-time denoising efficiency of the laser point cloud data.

[0138] The technical features of the above embodiments can be combined arbitrarily. For the sake of brevity, not all combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

Claims

1. A real-time denoising method for laser point cloud data, characterized in that, include: Multiple laser points are collected based on the detection of the arc-shaped cover by lidar, and the laser point cloud data is determined based on the synthesis of multiple laser points; Multiple point cloud regions are determined based on the division of laser point cloud data, and multiple sub-anomaly regions are determined based on the region detection of each point cloud region, and the corresponding anomaly locations are marked. Based on the actual image of the arc-shaped cover, multiple sub-anomaly regions and corresponding anomaly locations, abnormal laser point cloud data is determined. Based on the morphology of the abnormal laser point cloud data and the surface contour of the arc-shaped cover, real-time denoising measures for the abnormal laser point cloud data are determined to optimize the laser point cloud data in an overall manner. The real-time denoising process of acquiring laser point cloud data is used to determine the factors affecting denoising efficiency based on the real-time denoising process of laser point cloud data. Based on the influencing factors, unprocessed abnormal laser point cloud data, and the data load of the lidar, optimization measures for real-time denoising are determined, and dynamic optimization of the real-time denoising process of the laser point cloud data is triggered to improve the real-time denoising efficiency of the laser point cloud data.

2. The real-time denoising method for laser point cloud data according to claim 1, characterized in that, The process of acquiring multiple laser points based on lidar detection of the arc-shaped cover, and determining laser point cloud data based on the synthesis of these multiple laser points, includes: When the lidar detects the curved cover, it collects multiple laser points corresponding to the curved cover and presents multiple laser points in the same coordinate system, with the multiple laser points distributed at different positions in the coordinate system. In the same coordinate system, multiple discrete laser points are identified based on the recognition of multiple laser points. The positions corresponding to the multiple discrete laser points are the outer peripheral positions of the arc-shaped cover. Laser point cloud data is determined based on the optimization of multiple discrete laser points. This laser point cloud data does not contain multiple discrete laser points.

3. The real-time denoising method for laser point cloud data according to claim 1, characterized in that, The process involves determining multiple point cloud regions based on the division of laser point cloud data, identifying multiple sub-anomaly regions based on region detection within each point cloud region, and marking the corresponding anomaly locations, including: In laser point cloud data, multiple point cloud concentration areas are determined based on the division of laser point cloud data, and multiple point cloud regions are determined based on the location of multiple point cloud concentration areas and the connection between two adjacent point cloud concentration areas. Simultaneous detection is performed on multiple point cloud regions. At this time, multiple abnormal point cloud features are determined based on the region detection of each point cloud region. Multiple sub-anomaly regions are determined based on the location of multiple anomalous point cloud features and the regional distribution map of the point cloud regions. The corresponding anomaly location is determined based on the location detection of multiple sub-anomaly regions.

4. The real-time denoising method for laser point cloud data according to claim 1, characterized in that, The process involves determining abnormal laser point cloud data based on the actual image of the arc-shaped cover, multiple sub-abnormal regions, and corresponding abnormal locations. Real-time denoising measures for the abnormal laser point cloud data are then determined based on its morphology and the surface contour of the arc-shaped cover, in order to perform overall optimization of the laser point cloud data. This includes: Real-time monitoring of the arc-shaped cover and acquisition of actual images of the arc-shaped cover; determination of multiple first-anomaly point cloud data based on the actual images of the arc-shaped cover and multiple sub-anomaly regions; Multiple second abnormal point cloud data are determined based on the actual image of the arc-shaped cover and the corresponding abnormal location. Abnormal laser point cloud data are determined based on the matching of multiple first abnormal point cloud data and multiple second abnormal point cloud data.

5. The real-time denoising method for laser point cloud data according to claim 4, characterized in that, The process of determining abnormal laser point cloud data based on the actual image of the arc-shaped cover, multiple sub-abnormal regions, and corresponding abnormal locations, and determining real-time denoising measures for the abnormal laser point cloud data based on the morphology of the abnormal laser point cloud data and the surface contour of the arc-shaped cover to optimize the laser point cloud data holistically, also includes: The model acquires the shape of abnormal laser point cloud data, determines the corresponding point cloud anomaly type by comparing the shape of the abnormal laser point cloud data with the surface contour of the arc-shaped cover, and determines the real-time denoising measures for the abnormal laser point cloud data based on the mapping relationship between the point cloud anomaly type and real-time denoising measures. The overall optimization of the laser point cloud data is triggered based on the execution of the real-time denoising measures for the abnormal laser point cloud data.

6. The real-time denoising method for laser point cloud data according to claim 1, characterized in that, The real-time denoising process for the acquired laser point cloud data determines the factors affecting denoising efficiency based on this process, including: The system monitors the execution of real-time denoising measures for abnormal laser point cloud data and marks the processed abnormal laser point cloud data. The real-time denoising process of the laser point cloud data is determined based on the detection of the processed abnormal laser point cloud data.

7. The real-time denoising method for laser point cloud data according to claim 6, characterized in that, The real-time denoising process for acquiring laser point cloud data, based on which factors affecting denoising efficiency are determined, also includes: In the real-time denoising process of laser point cloud data, the denoising items to be processed and the denoising items being processed are determined based on the analysis of the real-time denoising process of laser point cloud data. The corresponding denoising efficiency range is determined based on the detection of the denoising items being processed. The first sub-influencing factor is determined based on the identification of the denoising efficiency range. The corresponding second sub-influencing factor is determined based on the denoising items to be processed. The influencing factors affecting the denoising efficiency are determined based on the combination of the first and second sub-influencing factors.

8. The real-time denoising method for laser point cloud data according to claim 1, characterized in that, The step of determining optimization measures for real-time denoising based on the influencing factors, unprocessed abnormal laser point cloud data, and the data load of the lidar, and triggering dynamic optimization of the real-time denoising process of the laser point cloud data to improve the real-time denoising efficiency of the laser point cloud data, includes: Collect various influencing factors, determine the first optimization coefficient based on the influencing factors and unprocessed abnormal laser point cloud data, determine the second optimization coefficient based on the influencing factors and the data load of the lidar, and determine the optimization measures for real-time denoising based on the mapping relationship between the first optimization coefficient, the second optimization coefficient and the optimization measures.

9. The real-time denoising method for laser point cloud data according to claim 8, characterized in that, The step of determining optimization measures for real-time denoising based on the influencing factors, unprocessed abnormal laser point cloud data, and the data load of the lidar, and triggering dynamic optimization of the real-time denoising process of the laser point cloud data to improve the real-time denoising efficiency of the laser point cloud data, further includes: The process of acquiring laser point cloud data in real time is used to determine the nodes to be optimized in the real time denoising process based on the real time denoising process and the unprocessed abnormal laser point cloud data. The nodes to be optimized are matched with the optimization measures of real time denoising. When the node to be optimized in the real-time denoising process is responded to, the optimization measures for real-time denoising are triggered, and the real-time denoising process of the laser point cloud data is dynamically optimized to gradually optimize the corresponding influencing factors and improve the real-time denoising efficiency of the laser point cloud data.

10. A real-time denoising system for laser point cloud data, characterized in that, The real-time laser point cloud data denoising system is applied to the real-time laser point cloud data denoising method as described in any one of claims 1-9, and the real-time laser point cloud data denoising system comprises: The laser point cloud data module is used to collect multiple laser points based on the detection of the arc-shaped cover by the lidar, and to determine the laser point cloud data based on the synthesis of multiple laser points; The sub-anomaly region module is used to determine multiple point cloud regions based on the division of laser point cloud data, determine multiple sub-anomaly regions based on the region detection of each point cloud region, and mark the corresponding anomaly locations. The real-time denoising module is used to determine abnormal laser point cloud data based on the actual image of the arc-shaped cover, multiple sub-abnormal regions and corresponding abnormal locations. Based on the shape of the abnormal laser point cloud data and the surface contour of the arc-shaped cover, it determines real-time denoising measures for the abnormal laser point cloud data in order to optimize the laser point cloud data as a whole. The influencing factors module is used to collect the real-time denoising process of laser point cloud data and determine the influencing factors that affect the denoising efficiency based on the real-time denoising process of laser point cloud data. The optimization module is used to determine optimization measures for real-time denoising based on the influencing factors, unprocessed abnormal laser point cloud data, and the data load of the lidar, and to trigger dynamic optimization of the real-time denoising process of the laser point cloud data to improve the real-time denoising efficiency of the laser point cloud data.

Images