A method and device for generating feature-enhanced maps for pipeline corridor scenes

Abstract

The present invention discloses a method and device for generating a feature-enhanced map for a tunnel scene, and belongs to the field of laser radar data processing. In existing point cloud map generation methods, in scenes such as underground tunnels, the point cloud map lacks obvious features due to excessive similarity of the surrounding environment, so it is easy to lose positioning information, and thus accurate positioning and navigation cannot be achieved. The present invention provides a method for generating a feature-enhanced map for a tunnel scene, which generates a feature-enhanced map for the tunnel scene by constructing a map clipping model, a feature extraction model, and a feature enhancement model, so that the map in the tunnel scene has obvious features, thereby effectively avoiding the loss of positioning information of the laser radar SLAM, and at the same time effectively improving the computational efficiency of the point cloud matching, thereby achieving accurate positioning and navigation of the tunnel scene, and ensuring that the robot or robot dog can perform normal inspections when working in the underground tunnel.

Description

Technical Field

[0001] The present invention relates to a method and device for generating a feature-enhanced map for a pipe gallery scene, and belongs to the field of laser radar data processing. Background Art

[0002] A Chinese patent application (publication number: CN114609649A) discloses a robot positioning and navigation method based on reinforcement learning lidar data. Road information is collected through the lidar body on the robot body. When collecting information, the noise reduction module in the robot body will filter the noise data to obtain valid data; the collected valid data is transmitted to the map construction module in the robot body, so that the valid data is integrated into the map construction module, thereby realizing real-time updating of the map construction module; after integrating the valid data, multiple paths are screened out through the navigation module in the robot body, and then the optimal path is given and the robot body is directed to set off. The remaining paths are the most alternative paths, thereby avoiding road obstacles.

[0003] The above solution uses lidar to collect road information to achieve robot positioning and navigation. However, in scenarios such as underground tunnels, the surrounding environment is too similar, resulting in a lack of obvious features in the point cloud map. This can easily lead to the loss of lidar SLAM positioning information, making it impossible to achieve accurate positioning and navigation, which in turn affects the normal inspection tasks of robots or robot dogs in underground tunnels.

[0004] The information disclosed in this Background Art is only for understanding the background of the present inventive concept and therefore it may include information that does not constitute prior art. Summary of the Invention

[0005] In response to the above problem or one of the above problems, an object of the present invention is to provide a method and device for generating a feature-enhanced map for a tunnel scene. By constructing a map clipping model, a feature extraction model, and a feature enhancement model, the original point cloud map is clipped, and the region growing segmentation algorithm is used to extract the rod features and plane features of the short-range map. The rod features and plane features of the long-range map are extracted based on the long-range segmentation mechanism of point cloud rasterization, so that the enhanced rod features can be obtained, and the enhanced rod features are spliced ​​with the plane features to generate a feature-enhanced map for the tunnel scene, so that the map in the tunnel scene has obvious features, thereby effectively avoiding the loss of positioning information of the lidar SLAM, and then realizing accurate positioning and navigation of the tunnel scene, ensuring that the robot or robot dog can patrol normally when working in the underground tunnel.

[0006] In response to the above problem or one of the above problems, the second purpose of the present invention is to provide a method and device for generating a feature-enhanced map for a pipeline corridor scene, which can effectively extract the obvious rod-shaped features on both sides of the pipeline corridor. Through a long-distance segmentation algorithm based on point cloud rasterization, a large-scale long corridor can be rasterized to the coordinate origin, effectively solving the disadvantage of the region growing segmentation algorithm being insensitive to long-distance point clouds, and realizing feature enhancement of a large-scale long corridor, thereby effectively reducing the risk of positioning loss and improving the success rate of point cloud matching. At the same time, the present invention processes the extracted rod-shaped features based on a filtering algorithm, filters out point clouds that are not conducive to positioning matching, retains more effective and obvious features, and thus effectively improves the computational efficiency of point cloud matching. The solution is scientific, reasonable, and feasible.

[0007] To achieve one of the above purposes, the first technical solution of the present invention is:

[0008] A method for generating a feature-enhanced map for a pipe gallery scene includes the following steps:

[0009] Step 1: Obtain the original point cloud map of the pipeline corridor scene;

[0010] Step 2: Using a pre-built map clipping model, the original point cloud map is clipped based on distance to obtain multiple local maps, which include at least a long-distance map and a short-distance map;

[0011] Step 3: Using a pre-built feature extraction model, a region growing segmentation algorithm is used to extract rod features and plane features for the near-range map, and a long-range segmentation mechanism based on point cloud rasterization is used to extract rod features and plane features for the long-range map.

[0012] Step 4: Use the pre-built feature enhancement model and the filtering algorithm to process the rod features to obtain enhanced rod features. The enhanced rod features are then spliced ​​with the plane features to generate a feature enhancement map for the pipeline corridor scene.

[0013] The present invention crops the original point cloud map by constructing a map clipping model, a feature extraction model, and a feature enhancement model, and uses a region growing segmentation algorithm to extract the rod features and plane features of the short-range map. The rod features and plane features of the long-range map are extracted based on a long-range segmentation mechanism of point cloud rasterization, thereby obtaining enhanced rod features. The enhanced rod features are then spliced ​​with the plane features to generate a feature-enhanced map for the tunnel scene, so that the map in the tunnel scene has obvious features. Therefore, the loss of positioning information of the laser radar SLAM can be effectively avoided, and accurate positioning and navigation of the tunnel scene can be realized, thereby ensuring that robots or robot dogs can perform normal inspections when working in underground tunnels.

[0014] Furthermore, the present invention can effectively extract the obvious rod-shaped features on both sides of the pipeline corridor. Through a long-distance segmentation algorithm based on point cloud rasterization, a large-scale long corridor can be rasterized to the coordinate origin, effectively solving the drawback of the region growing segmentation algorithm's insensitivity to long-distance point clouds, and realizing feature enhancement of large-scale long corridors, thereby effectively reducing the risk of positioning loss and improving the success rate of point cloud matching. At the same time, the present invention processes the extracted rod-shaped features based on a filtering algorithm, filters out point clouds that are not conducive to positioning matching, and retains more effective and obvious features, thereby effectively improving the computational efficiency of point cloud matching. The solution is scientific, reasonable, and feasible.

[0015] Furthermore, the number of long-distance maps can be one or more; the number of short-distance maps can be one or more.

[0016] The rod-shaped feature is a metal rod feature or a plastic rod feature.

[0017] As preferred technical measures:

[0018] Step 2: Use the pre-built map clipping model to clip the original point cloud map based on distance to obtain multiple local maps as follows:

[0019] According to the region growing segmentation algorithm, set the recognition distance threshold;

[0020] Based on the pipeline corridor scenario, obtain the pipeline corridor length;

[0021] According to the recognition distance threshold, the length of the pipeline corridor is evenly segmented to obtain segmentation information;

[0022] Based on the segmentation information, the original point cloud map is cropped to obtain multiple local maps.

[0023] As preferred technical measures:

[0024] Step 3: Using the pre-built feature extraction model, the region growing segmentation algorithm is used to extract the rod features and plane features of the close-range map as follows:

[0025] Step 31, obtaining point cloud data of the short-range map, which includes three-dimensional coordinate information of several spatial points;

[0026] Step 32: Calculate the normal vector and curvature value of each spatial point based on the three-dimensional coordinate information;

[0027] Based on the curvature value, the spatial points are sorted to obtain a spatial point array, which includes several spatial points and corresponding curvature values;

[0028] Step 33: Use the region growing segmentation algorithm to find the minimum curvature value from the spatial point array, obtain the spatial point corresponding to the minimum curvature value, and use the spatial point as the initial pipeline gallery seed point;

[0029] Step 34: Search for a neighborhood point set of the initial gallery seed point, which includes several neighborhood points. Perform a similarity criterion judgment on each neighborhood point and the initial gallery seed point. If the similarity criterion is met, merge it into the queue where the gallery seed point is located to obtain a gallery point cluster queue.

[0030] Step 35: for the remaining unmarked spatial points, loop through steps 33 and 34 to obtain several pipeline corridor point cluster queues;

[0031] Step 36: Process several pipeline corridor point cluster queues to obtain rod features and plane features.

[0032] As preferred technical measures:

[0033] The method for processing several pipeline corridor point clusters to obtain rod-shaped features and plane features is as follows:

[0034] Set the minimum and maximum number of point clusters for plane feature detection; and based on the minimum and maximum number of point clusters, screen several corridor point cluster queues to obtain non-planar queues and planar queues;

[0035] Calculate the normal vector angle or curvature difference of spatial points in a non-planar array;

[0036] When the normal vector angle or curvature difference is greater than the corresponding threshold, the spatial point is a non-planar point;

[0037] When the normal vector angle or curvature difference is less than or equal to the corresponding threshold, the space point is a plane point;

[0038] The non-planar points are aggregated to obtain the rod-shaped features;

[0039] Summarize the plane points and plane queues to obtain plane features;

[0040] Or / and, the method to calculate the normal vector and curvature value of each space point is as follows:

[0041] Obtain the neighboring point group and the number of neighboring points of a certain spatial point through principal component analysis;

[0042] Based on the neighbor point group and the number of neighbor points, calculate the mean point of the neighbor point group;

[0043] According to the mean point of the neighboring point group and the number of neighboring points, the neighborhood covariance matrix of the spatial point is constructed;

[0044] Then the eigenvector corresponding to the minimum eigenvalue of the covariance matrix is ​​used as the normal vector of the spatial point. The normal vector is used to characterize the local surface orientation of the point cloud.

[0045] Based on the eigenvalues ​​of the covariance matrix, the curvature value of the spatial point is calculated. The curvature value is used to characterize the degree of curvature of the local surface of the point cloud.

[0046] As preferred technical measures:

[0047] The method for judging the similarity criterion between each neighborhood point and the initial pipeline gallery seed point is as follows:

[0048] Similarity criteria include normal vector angle criterion and curvature difference criterion;

[0049] Obtain the normal vector of the neighborhood point and the normal vector of the initial gallery seed point, and calculate the angle between the normal vectors of the neighborhood point and the initial gallery seed point;

[0050] Compare the normal vector angle with the normal vector angle threshold to obtain an angle comparison result;

[0051] Obtain the curvature value of the neighborhood point and the curvature value of the initial gallery seed point, and calculate the curvature difference between the neighborhood point and the initial gallery seed point;

[0052] Compare the curvature difference with the curvature threshold to obtain the curvature comparison result;

[0053] Based on the angle comparison results and curvature comparison results, it is determined whether the neighborhood point and the initial pipeline gallery seed point belong to the same feature area; when the normal vector angle and curvature differences between the neighborhood point and the initial pipeline gallery seed point are both less than the corresponding thresholds, the neighborhood point and the initial pipeline gallery seed point belong to the same feature area.

[0054] As preferred technical measures:

[0055] Step 3: Using the pre-built feature extraction model, the method for extracting rod features and plane features of the long-range map based on the long-range segmentation mechanism of point cloud rasterization is as follows:

[0056] The first step is to obtain a long-distance map, rasterize and partition it into several grids, and record the minimum coordinates of each grid;

[0057] The second step is to use the minimum coordinate of the grid as the origin and translate the point cloud in the grid to obtain the grid map after translation transformation, so that each grid map is in a close range during region growing segmentation.

[0058] In the third step, each transformed raster map is segmented by region growing to extract plane features and rod features;

[0059] The fourth step is to perform reverse translation transformation on the plane features and rod features to make them consistent with the original long-distance map, thereby obtaining the rod features and plane features of the long-distance map.

[0060] As preferred technical measures:

[0061] The method of rasterizing and partitioning a long-distance map into several grids is as follows:

[0062] Get the extreme coordinates of the point cloud cluster in the long-distance map, including the minimum X-axis coordinate value, the maximum X-axis coordinate value, the maximum Y-axis coordinate value, and the minimum Y-axis coordinate value;

[0063] Set the X-axis resolution and Y-axis resolution of the grid;

[0064] Calculate the X-axis size and Y-axis size of the grid based on the X-axis resolution, Y-axis resolution and the extreme coordinates of the point cloud cluster;

[0065] Create an empty grid matrix based on the X-axis size and Y-axis size;

[0066] Based on the empty grid matrix, calculate the row index and column index of the point cloud;

[0067] According to the row index and column index of the point cloud, the point cloud is projected into the corresponding grid to obtain several grids, thereby realizing the raster partitioning of the long-distance map.

[0068] As preferred technical measures:

[0069] Step 4: Use the pre-built feature enhancement model and the filtering algorithm to process the rod features to obtain enhanced rod features. The enhanced rod features are then combined with the plane features to generate a feature enhancement map for the pipe gallery scene. The method is as follows:

[0070] Based on the filtering algorithm, outlier filtering is performed on the rod-shaped feature to remove outliers in the rod-shaped feature and obtain a new rod-shaped feature;

[0071] In order to retain more rod-shaped features, different voxel filtering downsampling parameters are set for the plane features and the new rod-shaped features, including the plane feature downsampling parameters and the rod feature downsampling parameters;

[0072] constructing a first downsampling filter based on the plane feature downsampling parameters;

[0073] constructing a second downsampling filter based on the rod-shaped object characteristic downsampling parameters;

[0074] Inputting the plane features into a first downsampling filter to obtain downsampled plane features;

[0075] Inputting the rod-shaped feature into a second downsampling filter to obtain enhanced rod-shaped feature;

[0076] Plane features and rod-shaped features are spliced ​​together to generate a feature-enhanced map for the pipeline corridor scene, thereby achieving map feature enhancement for the pipeline corridor scene.

[0077] As preferred technical measures:

[0078] The method for performing outlier filtering on the rod-shaped feature to remove outliers in the rod-shaped feature and obtain a new rod-shaped feature is as follows:

[0079] Process the rod-shaped object features to obtain several coordinate points;

[0080] For each coordinate point, calculate the Euclidean distance between it and its adjacent coordinate points;

[0081] Construct a spherical neighborhood based on Euclidean distance and search radius;

[0082] Evaluate the clustering density of the spherical neighborhood and calculate the number of inliers in the spherical neighborhood;

[0083] Based on the number of inliers and the inlier threshold, determine whether to retain the coordinate point. If the number of inliers is greater than or equal to the inlier threshold, the coordinate point is retained without filtering. If the number of inliers is less than the inlier threshold, the coordinate point is discarded and needs to be filtered to obtain a new rod-shaped feature.

[0084] To achieve one of the above purposes, the second technical solution of the present invention is:

[0085] A device comprising:

[0086] one or more processors;

[0087] a storage device for storing one or more programs;

[0088] When the one or more programs are executed by the one or more processors, the one or more processors implement the above-mentioned method for generating a feature-enhanced map for a pipeline corridor scene.

[0089] Compared with the existing technical solutions, the present invention has the following beneficial effects:

[0090] The present invention crops the original point cloud map by constructing a map clipping model, a feature extraction model, and a feature enhancement model, and uses a region growing segmentation algorithm to extract the rod features and plane features of the short-range map. The rod features and plane features of the long-range map are extracted based on a long-range segmentation mechanism of point cloud rasterization, thereby obtaining enhanced rod features. The enhanced rod features are then spliced ​​with the plane features to generate a feature-enhanced map for the tunnel scene, so that the map in the tunnel scene has obvious features. Therefore, the loss of positioning information of the laser radar SLAM can be effectively avoided, and accurate positioning and navigation of the tunnel scene can be realized, thereby ensuring that robots or robot dogs can perform normal inspections when working in underground tunnels.

[0091] Furthermore, the present invention can effectively extract the obvious rod-shaped features on both sides of the pipeline corridor. Through a long-distance segmentation algorithm based on point cloud rasterization, a large-scale long corridor can be rasterized to the coordinate origin, effectively solving the drawback of the region growing segmentation algorithm's insensitivity to long-distance point clouds, and realizing feature enhancement of large-scale long corridors, thereby effectively reducing the risk of positioning loss and improving the success rate of point cloud matching. At the same time, the present invention processes the extracted rod-shaped features based on a filtering algorithm, filters out point clouds that are not conducive to positioning matching, and retains more effective and obvious features, thereby effectively improving the computational efficiency of point cloud matching. The solution is scientific, reasonable, and feasible. BRIEF DESCRIPTION OF THE DRAWINGS

[0092] Figure 1 A schematic diagram of a flow chart of a method for generating a feature-enhanced map for a pipe gallery scene according to the present invention;

[0093] Figure 2 Another flowchart of the method for generating a feature-enhanced map for a pipe gallery scene according to the present invention is shown;

[0094] Figure 3 A schematic diagram of a local plane feature after segmentation according to the present invention;

[0095] Figure 4 A schematic diagram of the characteristics of a local metal rod after segmentation according to the present invention;

[0096] Figure 5 This is a schematic diagram of the metal rod features after point cloud rasterization and coordinate conversion in the present invention;

[0097] Figure 6 A feature enhancement map generated by applying the present invention. DETAILED DESCRIPTION

[0098] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0099] Rather, the present invention encompasses any alternatives, modifications, equivalents, and solutions that fall within the spirit and scope of the present invention as defined by the claims. Furthermore, to facilitate a better understanding of the present invention, certain specific details are described in detail below in the detailed description of the present invention. Those skilled in the art will be able to fully understand the present invention without these details.

[0100] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one skilled in the art to which this invention pertains. The terms used herein are for the purpose of describing specific embodiments only and are not intended to limit the present invention. As used herein, the term "or / and" includes any and all combinations of one or more of the associated listed items.

[0101] like Figure 1 As shown, the first specific embodiment of the feature enhancement map generation method for the pipeline corridor scene of the present invention is as follows:

[0102] A method for generating a feature-enhanced map for a pipe gallery scene includes the following steps:

[0103] Step 1: Obtain the original point cloud map of the pipeline corridor scene;

[0104] Step 2: Using a pre-built map clipping model, the original point cloud map is clipped based on distance to obtain multiple local maps, which include at least a long-distance map and a short-distance map;

[0105] Step 3: Using a pre-built feature extraction model, a region growing segmentation algorithm is used to extract rod features and plane features for the near-range map, and a long-range segmentation mechanism based on point cloud rasterization is used to extract rod features and plane features for the long-range map.

[0106] Step 4: Use the pre-built feature enhancement model and the filtering algorithm to process the rod features to obtain enhanced rod features. The enhanced rod features are then spliced ​​with the plane features to generate a feature enhancement map for the pipeline corridor scene.

[0107] The second specific embodiment of the feature enhancement map generation method for the pipeline corridor scene of the present invention is as follows:

[0108] A method for generating a feature-enhanced map for a pipe gallery scene includes the following steps:

[0109] Step 1: Obtain the original point cloud map of the pipeline corridor degradation scene through the pre-built map acquisition model;

[0110] Step 2: Using a pre-built map clipping model, the original point cloud map is clipped based on distance to obtain multiple local maps, which include at least a long-distance map and a short-distance map;

[0111] Step 3: Using the pre-built feature extraction model, a region growing segmentation algorithm is used to extract the metal pole features of the short-range map, and a long-range segmentation mechanism based on point cloud rasterization is used to extract the metal pole features of the long-range map;

[0112] Step 4: Use the pre-built feature enhancement model to filter and downsample the segmented metal rod features to obtain enhanced feature information about the pipeline corridor scene and achieve map feature enhancement.

[0113] The present invention can effectively extract rod-shaped metal objects on both sides of the large pipeline corridor scene, remove outliers, strengthen and highlight features, and compress redundant point cloud data in the map to retain more effective and obvious features. At the same time, the algorithm itself has strong versatility, easy operation, and low requirements for hardware resources. Using this map for positioning can effectively reduce the computing power consumption of map matching and improve the stability of positioning.

[0114] like Figure 2 As shown, the third specific embodiment of the method for generating a feature-enhanced map for a pipe gallery scene according to the present invention is as follows:

[0115] A feature-enhanced map generation method for pipeline corridor scenes is a map segmentation algorithm based on point cloud feature enhancement. It improves on the region growing segmentation algorithm and solves the problem that traditional algorithms are sensitive to parameters and have difficulty in accurately segmenting obvious features as the pipeline corridor scene gradually increases.

[0116] In this embodiment, the region growing segmentation algorithm includes the following contents:

[0117] The region growing algorithm divides the map by merging adjacent point cloud regions with similar features. The core idea is to start from the pipeline gallery seed point and gradually expand the region until the stopping condition is met. This example focuses on the selection of pipeline gallery seed points and the determination of the similarity criterion. The steps of the region growing segmentation algorithm for segmenting features are as follows:

[0118] The first step is to obtain point cloud data. This data is stored in the PCD file format, primarily recording the 3D spatial structure (xyz) information of each point. A k-nearest neighbor search is then performed on the point cloud to calculate its normal vector and curvature. The normal vector reflects the local surface orientation of the point cloud, while the curvature describes the degree of curvature of the local surface. Points are then sorted according to their curvature values.

[0119] The normal vector is calculated by principal component analysis (PCA), first constructing the midpoint of the point cloud The neighborhood covariance matrix of is calculated as follows:

[0120]

[0121] in, Indicates a point The covariance matrix of Indicates a point The number of neighboring points of represents the i-th neighbor point, represents the mean point of the neighboring point group, is the transpose symbol.

[0122] Then the covariance matrix The eigenvector corresponding to the minimum eigenvalue is used as the normal vector of the point .

[0123] curvature From the eigenvalues ​​of the above covariance matrix The calculation formula is as follows:

[0124]

[0125]

[0126] The second step is to select the seed point of the pipeline gallery, and use the point with the smallest point cloud curvature as the initial pipeline gallery seed point , which is expressed as follows:

[0127]

[0128] in, Represents the point cloud collection of the entire map, represents a point in the point cloud set P, Indicates the curvature value at this point.

[0129] The third step is the region growing process, adding the pipeline gallery seed points to the queue and marked as visited, searching for the neighboring point set of the seed point of the tunnel , which includes several neighborhood points, for each neighborhood point The similarity criterion is used to judge the current pipeline gallery seed point. If If it has not been visited and meets the similarity criteria, it will be merged into the queue where the gallery seed point is located. , marked as visited; if If the similarity criteria are not met, they are merged into the enhanced feature queue.

[0130] The similarity criteria include the normal vector angle criterion and the curvature difference criterion. When the normal vector angle and the curvature difference are both less than the specified threshold, it can be determined that the two points belong to the same feature area.

[0131] The expression of the normal vector angle criterion is as follows:

[0132]

[0133] in, is the normal vector angle, represents the normal vector of point p, represents the normal vector of point q, Indicates the angle threshold that the normal vector angle meets, .

[0134] The expression of the curvature difference criterion is as follows:

[0135]

[0136] in, represents the curvature of point p, represents the curvature of point q, represents the difference threshold that the curvature satisfies, .

[0137] In the fourth step, the second and third steps are repeated for the remaining unmarked neighborhood points until the termination condition is met. The termination condition is that the original point cloud queue is empty or the area size reaches the preset limit.

[0138] Set the minimum number of point clusters for plane feature detection to min (min=50) and the maximum number of point clusters to max (max=1000000). This embodiment will detect all points between min and max. Points that do not meet the plane feature, that is, points whose normal vector angle or curvature difference is greater than the corresponding threshold, are non-plane points. Non-plane points are points on the segmented rod-shaped metal objects. In this embodiment, the cable meets the plane feature requirements and will not be recognized. The map effect after region growing segmentation is as follows: Figure 3 、 Figure 4 As shown, Figure 3 The colored areas in the figure are the plane points after segmentation. Figure 4 The red area in the figure is the metal rod feature after segmentation.

[0139] In this embodiment, a point cloud map is defined by a coordinate origin (a point at the coordinates [0,0,0]), XYZ coordinate axes, and a 1:1 scale with the real world. When the point cloud map is located far from the origin, problems such as uneven point cloud density and difficulty in neighborhood search can arise, leading to drawbacks such as plane fitting failure and inaccurate normal curvature estimation. Therefore, the point cloud map needs to be segmented to produce a near-range map and a far-range map. The far-range map represents a map formed by a first point cloud set, consisting of all points in the original point cloud map that are more than L meters from the coordinate origin. The near-range map represents a map formed by a second point cloud set, consisting of all points in the original point cloud map that are no more than L meters from the coordinate origin. Both the far-range and near-range maps share a common coordinate origin, XYZ coordinate axes, and scale. L can be 10, 20, 30, 40, 45, or 50, and those skilled in the art can adjust it based on the actual scenario and hardware characteristics.

[0140] In this embodiment, a long-distance segmentation mechanism based on point cloud rasterization is used to process a 180-meter pipeline corridor scene, which includes the following:

[0141] The 180m map of the tunnel is cut into four equal-distance parts and saved. For the long-distance map, the simple region growing segmentation algorithm cannot properly segment the metal rod features of the tunnel. Therefore, this embodiment proposes a long-distance segmentation method based on point cloud rasterization to accurately estimate the plane. The steps of this method are as follows:

[0142] The first step is to rasterize the point cloud map according to different resolutions in the x and y directions, and record the minimum horizontal and vertical coordinates of each grid. The rasterization process is as follows:

[0143] S1, set the X-axis resolution of the grid , Y-axis resolution .

[0144] S2, divide the grid cells according to the resolution and calculate the size of the grid X axis , Y-axis size , which is expressed as follows:

[0145]

[0146] in, 、 、 、 They represent the minimum X-axis coordinate value, maximum X-axis coordinate value, minimum Y-axis coordinate value, and maximum Y-axis coordinate value of the point cloud cluster (grid). This is a function that rounds floating-point decimals upwards in the C++ standard.

[0147] S3, based on grid size 、 , create an empty grid matrix, the expression is as follows:

[0148]

[0149] in, Built to the C++ standard The function of the zero matrix of of OK A matrix with all zero columns.

[0150] S4, traversal point Project to the corresponding grid Among them are the row index and column index of the grid where the point cloud is located, respectively, which are calculated by the following formula:

[0151]

[0152] in, This is a function that rounds floating-point decimals down in the C++ standard.

[0153] After the above steps, the point cloud map rasterization operation is completed and the point cloud is stored in raster form.

[0154] The second step is to traverse the grid map in turn, perform translation transformation on the grid map, and transform the point cloud in each grid into the corresponding The calculation formula for translation based on the grid origin is:

[0155]

[0156] in, A point in the grid The coordinate value of yes The coordinate value after translation transformation based on the grid origin.

[0157] The third step is to perform region growing segmentation on each transformed grid map separately, which can ensure that each transformed grid map is in a close distance state during region growing segmentation, so that the plane and rod features can be accurately segmented;

[0158] The fourth step is to save the segmented plane and obvious features and perform reverse translation transformation to make it consistent with the original Figure 1Finally, the segmented planes and obvious features are merged to generate a new point cloud map. The point cloud map segmentation results after point cloud rasterization can be seen in Figure 5 .

[0159] In this embodiment, the method for filtering and downsampling regional outliers includes the following:

[0160] Since the segmented metal rod features contain a large number of outliers and noise points, while the plane points, as features with less influence in pipe corridor positioning, have almost no outliers, outlier filtering is introduced into the segmented metal rod features. Based on density clustering, the point cloud is divided into several clusters. Each cluster meets the minimum number of points and neighborhood radius requirements, and isolated points that do not belong to any cluster are eliminated. The specific process is as follows:

[0161] Step 1: For each point in the metal rod feature , search its spherical neighborhood , which is calculated as follows:

[0162]

[0163] in, is the search radius, in units , yes and The Euclidean distance of .

[0164] Step 2: For the spherical neighborhood The cluster density is evaluated and the number of points in the neighborhood is calculated. , including the center point itself, is calculated as follows:

[0165]

[0166] Step 3: Based on the inlier threshold Determine whether to keep the point. If , then keep the point without filtering, if , then the point is removed and filtering is required for the point.

[0167] Since there are a large number of planar points in the original point cloud map, these planar points are not very helpful for positioning. On the contrary, too many planar points will consume more time for feature matching, thus affecting the real-time performance of positioning. Therefore, in order to retain more metal rod features, different voxel filtering downsampling parameters are set for the planar features and sharp metal rod features after segmentation. The method is as follows:

[0168] (1) Set the downsampling parameter of the segmented plane feature to 0.15 to obtain the first downsampling filter;

[0169] (2) Set the metal rod feature downsampling parameter to 0.05 to obtain the second downsampling filter;

[0170] (3) The segmented features are input into different downsampling filters to obtain downsampled plane features and metal rod features;

[0171] (4) The plane features and metal rod features are combined into a complete point cloud map to further enhance the rod features. The final feature enhancement map can be found in Figure 6 .

[0172] Furthermore, by counting the number of point clouds before and after feature enhancement, we found that the original map had 2,353,341 point clouds, and the segmented map had 229,389 point clouds. Therefore, the proportion of retained point clouds is 9.7%, indicating that the present invention can correctly segment prominent features in the pipeline corridor environment map and enhance effective features. The enhanced point cloud map was applied to actual pipeline corridor inspection scenarios. Long-term testing showed that the map can effectively reduce feature matching time, save computing resources, and improve positioning stability.

[0173] Therefore, the present invention can effectively extract the obvious rod-shaped features on both sides of the pipeline corridor. Through the long-distance segmentation algorithm based on point cloud rasterization, the long corridor in a large area can be rasterized to the coordinate origin, effectively solving the disadvantage of the regional growth segmentation algorithm being insensitive to long-distance point clouds, and realizing feature enhancement of the long corridor in a large area. At the same time, the point cloud processing algorithm of regional outlier filtering and downsampling is used to retain more effective obvious features, filter out point clouds that are not conducive to positioning matching, and improve the computational efficiency of point cloud matching. Finally, the processed point cloud map is used for the 1.5-kilometer pipeline corridor scene on site. After experimental verification, the present invention can effectively improve the success rate of point cloud matching, improve the computational efficiency of the positioning algorithm, and reduce the risk of positioning loss.

[0174] An embodiment of a device applying the method of the present invention:

[0175] An electronic device comprising:

[0176] one or more processors;

[0177] a storage device for storing one or more programs;

[0178] When the one or more programs are executed by the one or more processors, the one or more processors implement the above-mentioned method for generating a feature-enhanced map for a pipeline corridor scene.

[0179] A computer medium embodiment of the method of the present invention:

[0180] A computer-readable storage medium stores a computer program, which, when executed by a processor, implements the above-mentioned method for generating a feature-enhanced map for a pipeline corridor scene.

[0181] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware. Furthermore, the present application may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, optical storage, etc.) containing computer-usable program code.

[0182] The model in this application is an object that objectively describes the morphological structure with the help of physical or virtual representation. The object is not equal to the physical body and is not limited to physical and virtual. It can be a data processing function, software program, processing mode, usage method, operation method, workflow, application process, electronic hardware, circuit module, processing system, system imitation or simulation object.

[0183] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the above embodiments, ordinary technicians in the field can still modify or replace the specific implementation methods of the present invention. Any modification or equivalent replacement that does not depart from the spirit and scope of the present invention should be included in the scope of protection of the claims of the present invention.

Claims

1. A method for generating a feature-enhanced map for a pipe gallery scene, characterized by: The following steps are involved: Step 1: Obtain the original point cloud map of the pipeline corridor scene; Step 2: Using a pre-built map clipping model, the original point cloud map is clipped based on distance to obtain multiple local maps, which include at least a long-distance map and a short-distance map; Step 3: Using a pre-built feature extraction model, a region growing segmentation algorithm is used to extract rod features and plane features for the near-range map, and a long-range segmentation mechanism based on point cloud rasterization is used to extract rod features and plane features for the long-range map. It includes the following: The first step is to obtain a long-distance map, rasterize and partition it into several grids, and record the minimum coordinates of each grid; The second step is to use the minimum coordinate of the grid as the origin and translate the point cloud in the grid to obtain the grid map after translation transformation, so that each grid map is in a close range during region growing segmentation. In the third step, each transformed raster map is segmented by region growing to extract plane features and rod features; The fourth step is to perform reverse translation transformation on the plane features and rod features to make them consistent with the original long-distance map, thus obtaining the rod features and plane features of the long-distance map; Step 4: Use the pre-built feature enhancement model and the filtering algorithm to process the rod features to obtain enhanced rod features. The enhanced rod features are then spliced ​​with the plane features to generate a feature enhancement map for the pipeline corridor scene.

2. A method for generating a feature-enhanced map for a utility corridor scene according to claim 1, characterized in that: Step 2: Use the pre-built map clipping model to clip the original point cloud map based on distance to obtain multiple local maps as follows: According to the region growing segmentation algorithm, set the recognition distance threshold; Based on the pipeline corridor scenario, obtain the pipeline corridor length; According to the recognition distance threshold, the length of the pipeline corridor is evenly segmented to obtain segmentation information; Based on the segmentation information, the original point cloud map is cropped to obtain multiple local maps.

3. The method for generating a feature-enhanced map for a utility corridor scene according to claim 1, wherein: Step 3: Using the pre-built feature extraction model, the region growing segmentation algorithm is used to extract the rod features and plane features of the close-range map as follows: Step 31, obtaining point cloud data of the short-range map, which includes three-dimensional coordinate information of several spatial points; Step 32: Calculate the normal vector and curvature value of each spatial point based on the three-dimensional coordinate information; Based on the curvature value, the spatial points are sorted to obtain a spatial point array, which includes several spatial points and corresponding curvature values; Step 33: Use the region growing segmentation algorithm to find the minimum curvature value from the spatial point array, obtain the spatial point corresponding to the minimum curvature value, and use the spatial point as the initial pipeline gallery seed point; Step 34: Search for a neighborhood point set of the initial gallery seed point, which includes several neighborhood points. Perform a similarity criterion judgment on each neighborhood point and the initial gallery seed point. If the similarity criterion is met, merge it into the queue where the gallery seed point is located to obtain a gallery point cluster queue. Step 35: for the remaining unmarked spatial points, loop through steps 33 and 34 to obtain several pipeline corridor point cluster queues; Step 36: Process several pipeline corridor point cluster queues to obtain rod features and plane features.

4. The method for generating a feature-enhanced map for a utility corridor scene according to claim 3, wherein: The method for processing several pipeline corridor point clusters to obtain rod-shaped features and plane features is as follows: Set the minimum and maximum number of point clusters for plane feature detection; and based on the minimum and maximum number of point clusters, screen several corridor point cluster queues to obtain non-planar queues and planar queues; Calculate the normal vector angle or curvature difference of spatial points in a non-planar array; When the normal vector angle or curvature difference is greater than the corresponding threshold, the spatial point is a non-planar point; When the normal vector angle or curvature difference is less than or equal to the corresponding threshold, the space point is a plane point; The non-planar points are aggregated to obtain the rod-shaped features; Summarize the plane points and plane queues to obtain plane features; Or / and, the method to calculate the normal vector and curvature value of each space point is as follows: Obtain the neighboring point group and the number of neighboring points of a certain spatial point through principal component analysis; Based on the neighbor point group and the number of neighbor points, calculate the mean point of the neighbor point group; According to the mean point of the neighboring point group and the number of neighboring points, the neighborhood covariance matrix of the spatial point is constructed; Then the eigenvector corresponding to the minimum eigenvalue of the covariance matrix is ​​used as the normal vector of the spatial point. The normal vector is used to characterize the local surface orientation of the point cloud. Based on the eigenvalues ​​of the covariance matrix, the curvature value of the spatial point is calculated. The curvature value is used to characterize the degree of curvature of the local surface of the point cloud.

5. The method for generating a feature-enhanced map for a pipeline corridor scene according to claim 3, wherein: The method for judging the similarity criterion between each neighborhood point and the initial pipeline gallery seed point is as follows: Similarity criteria include normal vector angle criterion and curvature difference criterion; Obtain the normal vector of the neighborhood point and the normal vector of the initial gallery seed point, and calculate the angle between the normal vectors of the neighborhood point and the initial gallery seed point; Compare the normal vector angle with the normal vector angle threshold to obtain an angle comparison result; Obtain the curvature value of the neighborhood point and the curvature value of the initial gallery seed point, and calculate the curvature difference between the neighborhood point and the initial gallery seed point; Compare the curvature difference with the curvature threshold to obtain the curvature comparison result; Based on the angle comparison results and curvature comparison results, determine whether the neighborhood point and the initial pipeline gallery seed point belong to the same feature area; When the difference in normal vector angle and curvature between the neighborhood point and the initial gallery seed point is less than the corresponding threshold, the neighborhood point and the initial gallery seed point belong to the same feature area.

6. The method for generating a feature-enhanced map for a utility corridor scene according to claim 1, wherein: The method of rasterizing and partitioning a long-distance map into several grids is as follows: Get the extreme coordinates of the point cloud cluster in the long-distance map, including the minimum X-axis coordinate value, the maximum X-axis coordinate value, the maximum Y-axis coordinate value, and the minimum Y-axis coordinate value; Set the X-axis resolution and Y-axis resolution of the grid; Calculate the X-axis size and Y-axis size of the grid based on the X-axis resolution, Y-axis resolution and the extreme coordinates of the point cloud cluster; Create an empty grid matrix based on the X-axis size and Y-axis size; Based on the empty grid matrix, calculate the row index and column index of the point cloud; According to the row index and column index of the point cloud, the point cloud is projected into the corresponding grid to obtain several grids, thereby realizing the raster partitioning of the long-distance map.

7. The method for generating a feature-enhanced map for a utility corridor scene according to claim 1, wherein: Step 4: Use the pre-built feature enhancement model and the filtering algorithm to process the rod features to obtain enhanced rod features. The enhanced rod features are then combined with the plane features to generate a feature enhancement map for the pipe gallery scene. The method is as follows: Based on the filtering algorithm, outlier filtering is performed on the rod-shaped feature to remove outliers in the rod-shaped feature and obtain a new rod-shaped feature; In order to retain more rod-shaped features, different voxel filtering downsampling parameters are set for the plane features and the new rod-shaped features, including the plane feature downsampling parameters and the rod feature downsampling parameters; constructing a first downsampling filter based on the plane feature downsampling parameters; constructing a second downsampling filter based on the rod-shaped object characteristic downsampling parameters; Inputting the plane features into a first downsampling filter to obtain downsampled plane features; Inputting the rod-shaped feature into a second downsampling filter to obtain enhanced rod-shaped feature; Plane features and rod-shaped features are spliced ​​together to generate a feature-enhanced map for the pipeline corridor scene, thereby achieving map feature enhancement for the pipeline corridor scene.

8. The method for generating a feature-enhanced map for a utility corridor scene according to claim 7, wherein: The method for performing outlier filtering on the rod-shaped feature to remove outliers in the rod-shaped feature and obtain a new rod-shaped feature is as follows: Process the rod-shaped object features to obtain several coordinate points; For each coordinate point, calculate the Euclidean distance between it and its adjacent coordinate points; Construct a spherical neighborhood based on Euclidean distance and search radius; Evaluate the clustering density of the spherical neighborhood and calculate the number of inliers in the spherical neighborhood; Based on the number of inliers and the inlier threshold, determine whether to retain the coordinate point. If the number of inliers is greater than or equal to the inlier threshold, the coordinate point is retained without filtering. If the number of inliers is less than the inlier threshold, the coordinate point is discarded and needs to be filtered to obtain a new rod-shaped feature.

9. A device, characterized in that: It includes: one or more processors; a storage device for storing one or more programs; When the one or more programs are executed by the one or more processors, the one or more processors implement a feature enhancement map generation method for a pipeline corridor scene as described in any one of claims 1 to 8.

Images