A multi-line laser radar (SLAM) point cloud densification method based on indoor environment features

Abstract

The application discloses a multi-line laser radar SLAM point cloud densification method based on indoor environment characteristics, and the method comprises the following steps: using first frame of rectified point cloud data as a reference point cloud to construct an initialization map; performing feature extraction, data matching and feature fusion on each frame of rectified point cloud data, and then constructing a residual equation; according to the residual equation, the non-linear least square method is used to iteratively optimize the residual of lines and planes, and finally the optimized pose is output; according to the optimized pose, the point cloud data of each frame on the current each scanning line is classified into plane points and corner points, the relationship between the plane points of the current frame and the surrounding neighborhood points is obtained, the straight line fitting of the plane points on the current frame is judged, and the point cloud densification processing is performed; the above method solves the problem that the multi-line laser radar data rapidly decreases in density far away from the radar, and through the intelligent densification algorithm, the uniform distribution of the point cloud at different depths is ensured, so that the quality of the entire indoor environment model is improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of point cloud data processing, and in particular to a multi-line laser radar SLAM point cloud densification method based on indoor environment features. BACKGROUND

[0002] In the development trend of modern automation and intelligence, indoor automatic navigation and environment modeling technology is particularly critical. Indoor environments often contain various geometric structures such as walls, columns, furniture, etc. These surface feature-rich structures pose higher requirements on laser SLAM technology. Multi-line laser radars (such as 16-line, 32-line, etc.) can generate a large amount of point cloud data due to their high-density data acquisition capability, providing rich information for environment perception. However, existing technologies still face many challenges in using multi-line laser radars for indoor environment point cloud densification, which is specifically manifested as follows:

[0003] On the one hand, the point cloud density is uneven: although the multi-line laser radar generates a large amount of point cloud data, the density distribution is usually uneven. For example, the density is higher near the center of the laser, while the density decreases rapidly away from the center of the laser, resulting in an inability to uniformly express surface information. This uneven point cloud distribution often leads to blurred and unclear details in the later color attachment and surface identification (such as furniture edges, decorations, etc.).

[0004] In addition, there is a problem of limited effect of existing encryption algorithms: most existing encryption algorithms use neighborhood point search methods to add new point clouds, but when dealing with surface edges (such as furniture edges, corners, etc.), the effect is often unsatisfactory. These algorithms are prone to produce noise at the edges or fail to accurately capture details, resulting in point clouds that deviate from the actual surface morphology. Some algorithms use plane information to encrypt point clouds, which perform well in flat areas, but in complex structure areas such as corners and bends, the effect is also poor. In addition, such methods have high computational complexity when processing large-scale point cloud data, resulting in slow processing speed and difficulty in meeting real-time application requirements.

[0005] The third aspect is the problem of unsatisfactory processing speed: existing point cloud densification algorithms often require high computational resources in pursuit of high density and accuracy, resulting in processing speed that cannot meet the real-time requirements of actual applications. Especially when the amount of point cloud data is large, the efficiency of the algorithm is particularly prominent, limiting its application in dynamic indoor environments.

[0006] Based on the above problems, how to utilize the high-density scanning characteristics of multi-line laser radars, combined with the feature information of laser SLAM generated point clouds, to achieve efficient and accurate indoor environment point cloud densification, has become a technical challenge that needs to be solved. SUMMARY

[0007] The application aims to provide a multi-line laser radar SLAM point cloud densification method based on indoor environment features, which solves the above technical problems pointed out in the prior art.

[0008] The application provides a multi-line laser radar SLAM point cloud densification method based on indoor environment features, comprising the following operation steps:

[0009] IMU data collected by an IMU sensor and point cloud data collected by a multi-line laser radar are acquired, timestamp analysis is performed on the data collected by the IMU sensor and the multi-line laser radar, and then timestamp alignment and synchronization are performed:

[0010] Data rectification is performed on single-frame point cloud data to obtain rectified single-frame point cloud data: using the rectified point cloud data of the first frame as a reference point cloud, an initialization map is constructed:

[0011] Feature extraction, data matching and feature fusion are performed on each frame of rectified point cloud data, and then a residual equation is constructed; according to the residual equation, the residual of the optimized line and plane is iteratively optimized by a nonlinear least square method, and the finally optimized pose is output

[0012] Each frame of point cloud data after matching optimization is distributed to the corresponding scanning line to realize scanning line classification; the point cloud data of each frame on the current scanning line is classified according to the optimized pose , into plane points and corner points, and plane point extraction and corner point extraction are performed:

[0013] The relationship between the plane points of the current frame and the surrounding neighborhood points is acquired, and the straight line fitting of the plane points on the current frame and the point cloud densification processing are judged to realize the output of the densified point cloud and the cyclic iterative optimization processing, and finally the high-precision high-density point cloud data and the dense point cloud map are output.

[0014] Preferably, the multi-line laser radar is a 16-line scanner or a 32-line scanner.

[0015] Preferably, the feature extraction, data matching and feature fusion are performed on each frame of rectified point cloud data, and then a residual equation is constructed; according to the residual equation, the residual of the optimized line and plane is iteratively optimized by a nonlinear least square method, and the finally optimized pose is output , specifically including:

[0016] For each frame of rectified point cloud data, line features and plane features are extracted therefrom;

[0017] The single-frame point cloud data is divided according to the beam ID (i.e. scanning line number), and the point cloud corresponding to each beam constitutes an independent processing unit.​

[0018] For the point cloud of each wire bundle, the local curvature of each point is calculated, and the line feature (high curvature) and the surface feature (low curvature) are distinguished by the curvature threshold; the points of each wire bundle are sorted by time, and then the n neighborhood points before and after the current point are searched by time, and the curvature is calculated by the following formula:

[0019] ;

[0020] Wherein, c is the calculated curvature, is the current point, is the time-continuous point before the current time, is the time-continuous point after the current time;

[0021] Set the curvature threshold T, if the curvature c is less than T, it is determined as a line feature, otherwise it is a surface feature; and the number of filtered line features and surface features is obtained by sorting according to the curvature c;

[0022] For the extracted line features and surface features, respectively, the neighborhood search matching is performed with the features in the initialized map, and the corresponding map feature line equation and surface equation are obtained;

[0023] The residual equation of the map feature line equation and the surface equation is constructed, so that the distance from the extracted plane point to the map surface plane is minimized, and the distance from the line feature point to the map line feature is minimized, wherein the line feature residual equation is constructed as:

[0024] ;

[0025] Wherein, is the extracted line parameter, is the optimized pose, is the extracted line feature;

[0026] The surface feature residual equation is constructed as:

[0027] ;

[0028] Wherein, is the extracted plane parameter, is the optimized pose, is the extracted surface feature;

[0029] According to the line feature residual and the surface feature residual, the nonlinear least squares method is used for iteration to obtain a residual set, which is expressed as:

[0030] ;

[0031] Wherein For the preset paradigm, by solving the function, the residual of the line and the plane is optimized to obtain the final optimized pose .

[0032] Preferably, the point cloud data of each frame after matching optimization is distributed to the corresponding scanning line to realize scanning line classification, specifically including:

[0033] The point cloud data of each frame after matching optimization is classified according to the line sequence in the point cloud attribute, so as to attribute the points in each frame of point cloud to different scanning lines of the laser scanner.

[0034] Preferably, the point cloud data of each frame on each scanning line is classified according to the optimized pose , into plane points and corner points, and plane point extraction and corner point extraction are performed, specifically including:

[0035] The point cloud on each scanning line is classified according to the optimized pose , into plane points and corner points:

[0036] If it is determined that the current point and other points on the scanning line are continuously increased in time, and the distance between adjacent points does not exceed a certain threshold, and the current point and other points on the scanning line are located on a straight line, and the searched neighborhood points of the current point can be fitted into a plane when the point cloud map is searched from the current point, it is determined that the neighborhood points of the current point are plane points;

[0037] If the current point and the surrounding points are not continuous in time, or the neighborhood points are not located on a straight line, and the finally searched points cannot be fitted into a plane, it is determined that the neighborhood points of the current point are corner points;

[0038] The formula for determining whether it is located on a straight line is as follows:

[0039] cos(A) = ;

[0040] Wherein, is the coordinate of the current point, is the coordinate of the previous time point on the current line, is the coordinate of the next time point on the current line, and A is the evaluation angle value of the current scanning point when determining whether the previous and next points are located on a straight line. If the angle value is close to 180°, it is closer to the straight line. If the angle value is less than 90°, it is considered to be a corner point;

[0041] The formula for determining whether the current point and its map search neighborhood points can be fitted into a plane is as follows:

[0042] ;

[0043] wherein , , , d is a plane equation to be solved, x, y, z is a neighborhood point in the map searched by the current point.

[0044] Preferably, the relationship of the plane point of the current frame and the surrounding neighborhood points is obtained, and it is judged to perform the line fitting of the plane point on the current frame and the point cloud densification processing, specifically including:

[0045] For the plane point on the current frame, it is judged whether the neighborhood points around it are plane points, if they are all plane points, then the straight line is fitted, if the straight line setting threshold is met, then uniform sampling is performed on the fitted straight line, and the points on the current fitted straight line are densified, and the fitting formula is as follows:

[0046] ;

[0047] wherein, is a newly added point, is a plane point on the current fitted straight line, is the main direction of the line, is the interval length of the added point;

[0048] If there is a corner point in the neighborhood of the plane point on the current frame, it is judged that it is close to the edge or turning point of the ground object, and no densification processing is performed, so as to reserve the sharp edge information and avoid edge blur.

[0049] Preferably, the densified point cloud output and the cyclic iteration optimization processing are realized, and finally the high-precision high-density point cloud data and the dense point cloud map are output, specifically including:

[0050] The current point cloud data after the densification processing is fused into the overall map to generate the final high-density and accurate dense point cloud map;

[0051] The sampled point cloud after the densification is processed, and it is judged whether there are overlapping points or redundant points, if it is determined that there are overlapping points or redundant points, the current overlapping points or redundant points are deleted.

[0052] The optimized high-density and accurate environment map point cloud is output after the overlapping points or redundant points are deleted.

[0053] Preferably, the single-frame point cloud data is numbered according to the beam ID, and the single-frame point cloud data is numbered according to the beam ID and the scanning line number

[0054] Correspondingly, the application provides a computer storage medium, comprising a memory for storing a computer program, a communication interface for connecting the memory with a processor, and the processor for executing the computer program to realize the indoor environment feature-based multi-line laser radar SLAM point cloud densification method.

[0055] Compared with the prior art, the embodiments of the application have at least the following technical advantages:

[0056] It can be seen from the above-mentioned indoor environment feature-based multi-line laser radar SLAM point cloud densification method provided by the application that, in specific application, firstly, IMU data collected by an IMU sensor and point cloud data collected by a multi-line laser radar are acquired, time stamp analysis is performed on the data collected by the IMU sensor and the multi-line laser radar, and then time stamp alignment and synchronization are performed.

[0057] Data rectification is performed on single-frame point cloud data to obtain rectified single-frame point cloud data: the first frame of rectified point cloud data is used as reference point cloud to construct an initialization map: feature extraction, data matching and feature fusion are performed on each frame of rectified point cloud data, and then a residual equation is constructed; according to the residual equation, the residual of the optimized line and plane is iteratively optimized by a nonlinear least square method, and the finally optimized pose is output. ;

[0058] Each frame of point cloud data after matching optimization is distributed to the corresponding scanning line to realize scanning line classification; each frame of point cloud data on the current scanning line is classified according to the optimized pose , into plane points and corner points, and plane point extraction and corner point extraction are performed.

[0059] The relationship between the plane points of the current frame and the surrounding neighborhood points is acquired, and the straight line fitting of the plane points on the current frame and the point cloud densification processing are judged to be performed; the densified point cloud output and the cyclic iteration optimization processing are realized, and finally the high-precision high-density point cloud data and the dense point cloud map are output.

[0060] The above technical solution realizes efficient and accurate point cloud densification by combining the SLAM technology and the features of the multi-line laser radar. On the one hand, it realizes efficient point cloud densification: by combining the high-density scanning features of the multi-line laser radar and the feature extraction of the SLAM technology, the application can effectively increase the density of the point cloud, especially in the expression of complex structures such as wall surfaces and furniture edges.

[0061] On the one hand, it improves the uniformity of point cloud data: it solves the problem of rapid density drop in multi-line lidar data at a distance from the radar, and through intelligent densification algorithm, it ensures uniform distribution of point cloud at different depths, thereby improving the quality of the entire indoor environment model.

[0062] On the one hand, it achieves adaptive surface feature densification: the densification algorithm considers the importance of surface features, giving higher weights to walls and complex structural areas, thus optimizing computational efficiency while preserving detail. Simultaneously, the algorithm design performs densification processing on the initial SLAM input simulation data, significantly improving processing speed and making it suitable for real-time navigation and environmental modeling in dynamic environments.

[0063] In summary, the present invention provides a multi-line lidar laser SLAM point cloud densification method based on indoor environmental features, which not only improves the efficiency and quality of point cloud densification processing, but also expands the application of lidar technology in indoor navigation and environmental modeling, providing a practical and efficient solution. Attached Figure Description

[0064] Figure 1 This is a schematic diagram of the main process of a multi-line lidar SLAM point cloud densification method based on indoor environmental characteristics in Example 1;

[0065] Figure 2 This is a flowchart of a specific operation in a multi-line lidar SLAM point cloud densification method based on indoor environmental characteristics, as described in Example 1.

[0066] Figure 3 This is another specific operation flowchart in the multi-line lidar SLAM point cloud densification method based on indoor environmental characteristics in Example 1;

[0067] Figure 4 This is a schematic diagram of a storage medium for applying the above-mentioned multi-line lidar SLAM point cloud densification method based on indoor environmental characteristics.

[0068] Labels: Processor 1110; Communication interface 1120; Memory 1130; Computer storage medium 1140. Detailed Implementation

[0069] The technical solution of the present invention will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0070] The present invention will now be described in further detail with reference to specific embodiments and accompanying drawings.

[0071] like Figure 1 As shown, this invention provides a method for densifying point clouds using multi-line lidar SLAM based on indoor environmental characteristics, comprising the following steps:

[0072] S101. Acquire IMU data collected by the IMU sensor and point cloud data collected by the multi-line LiDAR, perform timestamp parsing on the data collected by the IMU sensor and the multi-line LiDAR, and then perform timestamp alignment and synchronization:

[0073] S102. Perform data correction on the single-frame point cloud data to obtain the corrected single-frame point cloud data:

[0074] S103. Using the corrected point cloud data from the first frame as the reference point cloud, construct the initial map:

[0075] S104. For each frame of corrected point cloud data, feature extraction, data matching, and feature fusion are performed, and then a residual equation is constructed. Based on the residual equation, the residuals of lines and surfaces are iteratively optimized using a nonlinear least squares method, and the final optimized pose is output. ;

[0076] S105. Assign the point cloud data of each frame after matching optimization to the corresponding scan line to achieve scan line classification;

[0077] S106. The point cloud data of each frame on each of the current scan lines is adjusted according to the optimized pose. The neighborhood points are classified into two types: planar points and corner points. Planar point extraction and corner point extraction are then performed.

[0078] S107. Obtain the relationship between the planar points in the current frame and the surrounding neighboring points, and determine to perform line fitting and point cloud densification processing on the planar points in the current frame.

[0079] S108, realizes the output of denser point cloud and iterative optimization processing, and finally outputs high-precision and high-density point cloud data and dense point cloud map.

[0080] In practical applications, it first acquires IMU data from the IMU sensor and point cloud data from the multi-line LiDAR, performs timestamp parsing on the data from the IMU sensor and the multi-line LiDAR, and then performs timestamp alignment and synchronization.

[0081] Perform data correction on single-frame point cloud data to obtain corrected single-frame point cloud data: Use the first frame of corrected point cloud data as the reference point cloud to construct an initial map: Perform feature extraction, data matching, and feature fusion on each frame of corrected point cloud data, and then construct residual equations; Based on the residual equations, iteratively optimize the residuals of lines and surfaces using a nonlinear least squares method, and output the final optimized pose. ;

[0082] The optimized point cloud data of each frame is assigned to the corresponding scan line to achieve scan line classification; the point cloud data of each frame on each current scan line is then analyzed according to the optimized pose. The neighborhood points are classified into two types: planar points and corner points. Planar point extraction and corner point extraction are then performed.

[0083] The algorithm obtains the relationship between the planar points in the current frame and the surrounding neighboring points, determines and performs line fitting of the planar points in the current frame and point cloud densification processing; realizes the output of densified point cloud and iterative optimization processing, and finally outputs high-precision and high-density point cloud data and dense point cloud map.

[0084] The above technical solution achieves efficient and accurate point cloud densification by combining SLAM technology and the features of multi-line LiDAR. On one hand, it achieves efficient point cloud densification: by combining the high-density scanning features of multi-line LiDAR with the feature extraction of SLAM technology, this invention can effectively increase the density of the point cloud, especially in the detailed representation of complex structures such as walls and furniture edges.

[0085] On the one hand, it improves the uniformity of point cloud data: it solves the problem of rapid density drop in multi-line lidar data at a distance from the radar, and through intelligent densification algorithm, it ensures uniform distribution of point cloud at different depths, thereby improving the quality of the entire indoor environment model.

[0086] On the one hand, it achieves adaptive surface feature densification: the densification algorithm considers the importance of surface features, giving higher weights to walls and complex structural areas, thus optimizing computational efficiency while preserving detail. Simultaneously, the algorithm design performs densification processing on the initial SLAM input simulation data, significantly improving processing speed and making it suitable for real-time navigation and environmental modeling in dynamic environments.

[0087] In summary, the present invention provides a multi-line lidar laser SLAM point cloud densification method based on indoor environmental features, which not only improves the efficiency and quality of point cloud densification processing, but also expands the application of lidar technology in indoor navigation and environmental modeling, providing a practical and efficient solution.

[0088] This invention provides a method for point cloud densification using multi-line LiDAR SLAM based on indoor environmental features. This method offers a complete approach to point cloud densification using multi-line LiDAR SLAM based on indoor environmental features. By combining IMU data and multi-line LiDAR point cloud data, and through timestamp parsing and synchronization, feature extraction, residual optimization, and point cloud densification, it achieves high-precision and high-density point cloud map output. This solves the problem of sparse point clouds in indoor environments while improving the accuracy and densification quality of the point cloud.

[0089] This invention provides a multi-line lidar SLAM point cloud densification method based on indoor environmental features. This method combines the high-density scanning characteristics of multi-line lidar with the key surface features of SLAM-generated point clouds to achieve efficient and accurate point cloud densification, thereby improving the quality and application effect of indoor environmental point cloud maps.

[0090] Preferably, the multi-line LiDAR is a 16-line scanner or a 32-line scanner. By specifying the multi-line LiDAR as a 16-line or 32-line scanner, its high-density scanning capability is utilized to improve the efficiency of point cloud densification, adapting to the different performance characteristics of multi-line LiDAR.

[0091] Preferably, feature extraction, data matching, and feature fusion are performed on each frame of corrected point cloud data, and then a residual equation is constructed. Based on the residual equation, the residuals of lines and surfaces are iteratively optimized using a nonlinear least squares method, and the final optimized pose is output. Specifically, it includes:

[0092] For each frame of point cloud data after correction, extract the line features and surface features respectively;

[0093] The point cloud data of a single frame is divided according to the line bundle ID (i.e., scan line number), and the point cloud corresponding to each line bundle constitutes an independent processing unit;

[0094] For the point cloud of each line bundle, calculate the local curvature of each point, and distinguish between line features (high curvature) and surface features (low curvature) using a curvature threshold. Sort the points of each line bundle by time, and then search for the n neighboring points before and after the current point by time, calculating its curvature using the following formula:

[0095] ;

[0096] Where c is the calculated curvature, For the current point, Points that are consecutive in time preceding the current time. Points that are consecutive in time after the current time;

[0097] Set a curvature threshold T. If the curvature c is less than T, it is determined to be a line feature; otherwise, it is a surface feature. The number of line features and surface features selected is obtained by sorting according to the curvature c.

[0098] For the extracted line features and surface features, neighborhood search matching is performed between them and the features in the initialized map to obtain their corresponding map feature line equations and surface equations.

[0099] Construct residual equations for map feature line equations and surface equations to minimize the distance from extracted planar points to the map surface, and simultaneously minimize the distance from line feature points to map line features. The line feature residual equation is constructed as follows:

[0100] ;

[0101] in, For the extracted line parameters, For the optimized pose, For extracted line features;

[0102] The surface feature residual equation is constructed as follows:

[0103] ;

[0104] in, For the extracted plane parameters, For the optimized pose, For extracted surface features;

[0105] Based on the linear and surface feature residuals, an iterative process using nonlinear least squares is employed to obtain a residual set, which is expressed as:

[0106] ;

[0107] in Using a predefined paradigm, the residuals of lines and surfaces are optimized by solving this function to obtain the final optimized pose. .

[0108] The above-described specific implementation method proposes a residual optimization method based on line and surface features. By extracting feature points through curvature calculation and constructing residual equations, combined with iterative optimization using nonlinear least squares method, the accuracy of pose optimization is significantly improved, providing a more reliable foundation for subsequent point cloud densification.

[0109] Preferably, the step of assigning the optimized point cloud data of each frame to the corresponding scan line to achieve scan line classification specifically includes:

[0110] The point cloud data of each frame after matching and optimization is classified according to the line order in the point cloud attributes, so as to assign the points in each frame of point cloud to the corresponding different scanning lines of the laser scanner.

[0111] Preferably, the point cloud data of each frame on each of the current scan lines is based on the optimized pose. The neighborhood points are classified into two types: planar points and corner points. Planar point extraction and corner point extraction are then performed, specifically including:

[0112] The point cloud along each scan line is adjusted according to the optimized pose. Neighborhood points are classified into two types: planar points and corner points.

[0113] Determine the temporal continuity and distance relationship between the current point and other points on the scan line. If it is determined that the current point and other points on the scan line are continuously increasing in time, and the distance between adjacent points does not exceed a certain threshold; and it is determined that the current point and other points on the scan line are on a straight line; and when the current point is used to search the point cloud map, the neighboring points it finds can be fitted into a plane, then the neighboring points of the current point are determined to be planar points.

[0114] If the current point and its surrounding points are not in a continuous time frame, or if the neighboring points are not on a straight line, or if the last point found cannot be fitted to a plane, then the neighboring points of the current point are determined to be corner points.

[0115] The formula for determining whether two objects lie on a straight line is as follows:

[0116] cos(A) = ;

[0117] in, The coordinates of the current point. The coordinates of the current online position at the previous moment. A represents the coordinates of the current point on the line at the next moment. A is the evaluation angle value used to determine whether the points before and after the current scanning point are on a straight line. If the angle value is close to 180°, it is closer to a straight line. If the angle value is less than 90°, it is considered an inflection point.

[0118] The formula for determining whether the current point and its map search neighborhood points can fit a plane is as follows:

[0119] ;

[0120] in , , , d is the equation of the plane to be determined. x, y, z represents the neighboring points in the map searched from the current point.

[0121] The above specific implementation method, based on the optimized pose, classifies the point cloud data in the scan line into neighborhood points, dividing the data into planar points and corner points. The feature extraction further enhances the expressive power of the point cloud, especially in the detailed processing of complex regions (such as edges or turning points).

[0122] Better, such as Figure 2 As shown, the relationship between the planar points in the current frame and their surrounding neighboring points is obtained, and the line fitting and point cloud densification processing of the planar points in the current frame are performed. Specifically, this includes:

[0123] S1071. For a planar point in the current frame, determine whether its surrounding neighboring points are also planar points. If they are all planar points, fit a straight line. If the line meets the set threshold, perform uniform sampling on the fitted line to denser the points on the currently fitted line. The fitting formula is as follows:

[0124] ;

[0125] in, For newly added points, For the plane points on the currently fitted line, The main direction of the line. To increase the spacing between points;

[0126] S1072. If there is an inflection point in the neighborhood of a planar point in the current frame, it is determined that it is close to the edge or turning point of the ground object, and no densification processing is performed to preserve its sharp edge information and avoid edge blurring.

[0127] Better, such as Figure 3 As shown, the process implements point cloud densification output and iterative optimization, ultimately outputting high-precision, high-density point cloud data and a dense point cloud map. Specifically, this includes:

[0128] S1081. Integrate the current point cloud data after densification into the overall map to generate the final high-density, accurate dense point cloud map.

[0129] S1082. Sample the densed point cloud to determine if there are overlapping or redundant points. If overlapping or redundant points are determined, delete the current overlapping or redundant points.

[0130] S1083, and then output the optimized high-density, accurate environmental map point cloud after removing overlapping or redundant points.

[0131] The above-described implementation significantly improves the density of the point cloud through linear fitting of planar points and point cloud densification, especially in wall and planar areas, resulting in more accurate detail representation. Simultaneously, it avoids redundant densification at inflection points, improving the rationality and efficiency of densification. By promptly deleting redundant or overlapping points in the dense point cloud, the quality of the final point cloud map is optimized, generating a high-density, high-precision dense point cloud map that meets the needs of indoor navigation and environmental modeling.

[0132] Preferably, the single-frame point cloud data is numbered according to the harness ID, and the single-frame point cloud data is numbered according to the harness ID, which is consistent with the scan line number.

[0133] Example 2

[0134] This second embodiment is based on the multi-line lidar SLAM point cloud densification method based on indoor environmental characteristics provided in the first embodiment of the invention, and also provides a computer storage medium 1140 (hereinafter referred to as the storage medium). Figure 4 The diagram shown is a schematic of a computer storage medium structure framework provided in Embodiment 3 of the present invention, which includes:

[0135] Memory 1130 is used to store computer programs;

[0136] The communication interface 1120 is used to connect the memory 1130 to the processor 1110;

[0137] Processor 1110 is configured to execute a computer program to implement an embodiment of a multi-line lidar SLAM point cloud densification method based on indoor environmental features, as disclosed in any combination of the above embodiments.

[0138] This invention provides a multi-line lidar SLAM point cloud densification method based on indoor environmental features. This method combines the high-density scanning characteristics of multi-line lidar with the key surface features of SLAM-generated point clouds to achieve efficient and accurate point cloud densification, thereby improving the quality and application effect of indoor environmental point cloud maps.

[0139] In summary, this invention presents a method for point cloud densification based on indoor environmental features using multi-line LiDAR SLAM. This method provides a complete approach to point cloud densification based on indoor environmental features, combining IMU data and multi-line LiDAR point cloud data. Through timestamp parsing and synchronization, feature extraction, residual optimization, and point cloud densification, it achieves high-precision and high-density point cloud map output. This solves the problem of sparse point clouds in indoor environments while improving the accuracy and densification quality of the point cloud.

[0140] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; those skilled in the art can modify the technical solutions described in the foregoing embodiments, or make equivalent substitutions for some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for multi-line lidar SLAM point cloud densification based on indoor environment features, the method comprising: The method comprises the following operation steps: Obtaining IMU data collected by an IMU sensor and point cloud data collected by a multi-line laser radar, performing timestamp analysis on the data collected by the IMU sensor and the multi-line laser radar, and then performing timestamp alignment and synchronization; Performing data rectification on single-frame point cloud data to obtain rectified single-frame point cloud data; using the rectified point cloud data of the first frame as a reference point cloud to construct an initial map; The point cloud data after rectification of each frame is subjected to feature extraction, data matching and feature fusion, and then a residual error equation is constructed; according to the residual error equation, the residual error of the optimized line and plane is realized through iteration by a nonlinear least square method, and finally the optimized pose is output ; The point cloud data of each frame after matching optimization is distributed to the corresponding scan line to realize scan line classification; the point cloud data of each frame on the current scan line is classified according to the optimized pose of each frame, neighborhood point classification is performed, and the neighborhood points are divided into two types of plane points and corner points, and plane point extraction and corner point extraction are performed. The relationship of the planar point of the current frame and the surrounding neighborhood points is acquired, and it is judged to perform the straight line fitting of the planar point on the current frame and the point cloud densification processing, specifically comprising: for the planar point on the current frame, it is judged whether the surrounding neighborhood points are planar points, if they are all planar points, a straight line is fitted, if the straight line setting threshold is satisfied, uniform sampling is performed on the fitted straight line, and the points on the current fitted straight line are densified, and the fitting formula is as follows: ; wherein, is the newly added point, is the planar point on the current fitted straight line, is the main direction of the line, is the interval length of the added point; if the current frame has a corner point in the neighborhood of the planar point, it is judged that it is close to the edge or turning point of the ground object, and the densification processing is not performed; the output of the densified point cloud and the cyclic iteration optimization processing are realized, and finally the high-precision and high-density point cloud data and the dense point cloud map are output.

2. The method of claim 1, wherein, The multi-line laser radar is a 16-line scanner or a 32-line scanner.

3. The method of claim 1, wherein, The point cloud data after deviation correction of each frame is subjected to feature extraction, data matching and feature fusion, and then a residual equation is constructed; according to the residual equation, the residual of the optimized line and plane is realized through iteration by a nonlinear least square method, and finally the optimized pose is output , and specifically comprises: For each frame of rectified point cloud data, line features and surface features are extracted therefrom respectively; The single-frame point cloud data is divided according to the beam ID, and the point cloud corresponding to each beam constitutes an independent processing unit; For the point cloud of each beam, the local curvature of each point is calculated, and the line features and the surface features are distinguished by a curvature threshold; the points of each beam are sorted by time, and then n neighborhood points before and after the current point are searched according to time, and the curvature is calculated by the following formula: ； where c is a calculated curvature, is a point in time before the current time, is a point in time successive to the current time, is a point in time successive to the current time; A curvature threshold T is set, and if the curvature c is less than T, the line feature is determined, otherwise the surface feature is determined; and the line features and the surface features are sorted according to the curvature c to obtain the number of filtered line features and surface features; For the extracted line features and surface features, neighborhood search matching is performed with the features in the initial map to obtain corresponding map feature line equations and surface equations; Residual equations of the map feature line equations and the surface equations are constructed so that the distance from the extracted plane points to the map plane is minimized, and the distance from the line feature points to the map line features is minimized, wherein the line feature residual equation is constructed as: ； wherein, is the extracted line parameter, is the optimized pose, is the extracted line feature; The surface feature residual equation is constructed as: ； wherein, is the extracted plane parameter, is the optimized pose, is the extracted face feature; According to the line feature residual and the surface feature residual, a nonlinear least squares method is used for iteration to obtain a residual set, which is expressed as: ； wherein is a preset paradigm, by solving the function, the residual of the line and the plane is optimized, and a final optimized pose is obtained .

4. The method of claim 3, wherein the method is based on indoor environment features. The matched and optimized point cloud data of each frame is allocated to the corresponding scanning line to realize scanning line classification, specifically including: The matched and optimized point cloud data of each frame is classified according to the line sequence in the point cloud attribute, so as to attribute the points in each frame of point cloud to different scanning lines of the corresponding laser scanner.

5. The method of claim 4, wherein, The point cloud data of each frame on the current each scan line is classified according to the optimized pose , the neighborhood point classification is performed, and the plane points and corner points are divided, and the plane point extraction and the corner point extraction are performed, specifically including: The point cloud on each scan line is classified into plane points and corner points according to the optimized pose , If it is determined that the current point and other points on the scanning line are continuously increased in time, and the distance between adjacent points does not exceed a certain threshold, and the current point and other points on the scanning line are located on a straight line, and the neighborhood points searched by the current point can be fitted into a plane, it is determined that the neighborhood points of the current point are plane points. If the current point and the surrounding points are not continuous in time, or the neighborhood points are not located on a straight line, and the last searched points cannot be fitted into a plane, it is determined that the neighborhood points of the current point are corner points. The formula for determining whether the current point and its map search neighborhood points can be fitted into a plane is as follows: cos(A) = ; wherein, is the coordinate of the current point, is the coordinate of the previous point on the current line at the previous time, is the coordinate of the next point on the current line at the next time, and A is an evaluation angle value for determining whether the previous and next points at the current scanning point time are located on a straight line. If the angle value is close to 180°, it is closer to a straight line. If the angle value is less than 90°, it is considered to be an inflection point. The formula for determining whether the current point and its map search neighborhood points can be fitted into a plane is as follows: ； wherein , , , Pd is the plane equation to be determined, P k x, P k y, P k z are the neighborhood points in the map searched by the current point.

6. The method of claim 1, wherein, The dense point cloud output and the cyclic iteration optimization processing are realized, and finally high-precision and high-density point cloud data and dense point cloud maps are output, specifically including: The current point cloud data after the densification processing is fused into the overall map to generate a final high-density and accurate dense point cloud map; The dense point cloud is subjected to sampling processing to determine whether there are overlapping points or redundant points, and if it is determined that there are overlapping points or redundant points, the current overlapping points or redundant points are deleted; Then, the optimized high-density and accurate environment map point cloud after the overlapping points or redundant points are deleted is output.

7. The method of claim 1, wherein, The single-frame point cloud data is numbered according to the beam ID, and the single-frame point cloud data is numbered according to the beam ID and the scanning line number.

8. A computer-readable storage medium having stored thereon a computer program, characterized in that, The computer program, when executed by a processor, implements the steps of the method of any one of claims 1-7.

Images