Volume measurement system for test pit based on three-dimensional laser scanning

Abstract

The 3D laser scanning-based test pit volume measurement system relates to the field of engineering surveying technology, primarily addressing the technical problems of low efficiency, poor accuracy, and significant susceptibility to human factors in traditional test pit volume measurement methods. Key technical features of the system include: employing a 3D laser scanner with a scanning distance of 0.5-150m and a point cloud density of 10,000 to 500,000 points per square meter to scan the test pit area from multiple sites; using a point cloud registration module in the control processing unit to identify four symmetrically arranged targets to achieve multi-site cloud data fusion; after noise reduction using a point cloud filtering module, a surface model is constructed by a triangulation modeling module; and finally, the closed volume is calculated by a volume calculation module based on the spatial relationship between the triangulation model and the fitted reference plane. This system is mainly used for accurate earthwork measurement in filling and excavation projects in fields such as water conservancy, transportation, and mining.

Description

Technical Field

[0001] This invention relates to the field of engineering measurement technology. More specifically, this invention relates to a test pit volume measurement system based on three-dimensional laser scanning. Background Technology

[0002] In the fields of engineering surveying and geological exploration, accurate measurement of test pit volume plays a crucial role in calculating earthwork volume, monitoring settlement deformation, and evaluating the quality of engineering construction. Traditional surveying methods mainly employ direct contact methods, such as the grid method and sand replacement method. These methods are not only cumbersome and inefficient, but also struggle to accurately capture the complex surface morphology of pit shapes. With the development of surveying technology, non-contact 3D laser scanning technology has been gradually applied to volume measurement due to its ability to rapidly acquire high-density 3D surface data. However, in practical engineering applications, determining the reference plane remains a key technical challenge. Existing methods typically rely on point cloud data from the edge of the test pit to fit and generate a reference plane, but this method has inherent limitations; the point cloud data at the pit opening edge is often missing or incomplete due to soil loosening, collapse, or vegetation obstruction. Therefore, obtaining a stable and reliable reference plane has become a key technical problem that needs to be solved in 3D laser scanning-based test pit volume measurement technology. Summary of the Invention

[0003] One objective of this invention is to provide a test pit volume measurement system based on three-dimensional laser scanning.

[0004] To achieve these objectives and other advantages of the present invention, a test pit volume measurement system based on three-dimensional laser scanning is provided, comprising: The 3D laser scanner has a scanning distance range of 0.5-150m and a point cloud density of 10,000 to 500,000 points per square meter. The control and processing unit has a built-in point cloud registration module, point cloud filtering module, triangulation modeling module and volume calculation module. The 3D laser scanner establishes a data connection with the control and processing unit through a wireless transmission module. Four targets are arranged symmetrically around the center of the test pit at four locations on the outer edge of the test pit. During measurement: The three-dimensional laser scanner is set up on a tripod and aligned with the test pit scanning area. The three-dimensional laser scanner acquires point cloud data containing the target and the inner surface of the test pit from at least two different scanning stations. The point cloud registration module identifies and matches the spatial coordinates of the target sphere centers of four targets in the point cloud data of different scanning stations, and registers the point cloud data to the same coordinate system. The point cloud filtering module uses a statistical filtering algorithm to denoise the registered point cloud data. The threshold range for determining the number of neighboring points in the statistical filtering algorithm is 5-50. The triangular mesh modeling module is used to construct a triangular mesh model from the denoised point cloud data; The volume calculation module calculates the volume of the enclosed space between the triangular mesh model and the initial reference plane, based on the spatial relationship between the triangular mesh model and the initial reference plane, which is generated by fitting the center coordinates of the target spheres of at least three non-collinear targets.

[0005] Preferably, the statistical filtering algorithm of the point cloud filtering module is an adaptive threshold statistical filtering algorithm; The adaptive threshold statistical filtering algorithm first calculates the overall point density of the registered point cloud, and then dynamically adjusts the threshold used to determine the number of neighboring points based on the overall point density. The higher the point density, the larger the threshold is.

[0006] Preferably, the point cloud filtering module also integrates a radius filtering algorithm; The execution logic of the point cloud filtering module is as follows: first, a statistical filtering algorithm is used for the first stage of denoising to remove discrete noise points; then, a radius filtering algorithm is used for the second stage of denoising. The search radius of the radius filtering is set to 2-3 times the average point spacing of the point cloud to further remove outlier noise clusters.

[0007] Preferably, the point cloud registration module is configured to execute a target-surface feature dual-constraint dynamic weighting algorithm, which includes the following steps: Based on the spatial coordinates of the target sphere centers of the four targets, the initial registration transformation parameters are calculated. Two to three sets of depth-curvature correlation features were extracted from the point cloud data of each scanning station to the inner surface of the test pit. The depth-curvature correlation features include the continuous concave contour at a specific depth of the pit wall and the slope change zone between the center and the edge of the pit bottom. A site spacing-feature weight mapping model is constructed: registration weights are dynamically allocated based on the spatial distance between adjacent scanning sites. When the site spacing is no greater than 2m, the target coordinate matching weight is set to 0.75 to 0.85, and the depth-curvature correlation feature matching weight is set to 0.15 to 0.25. When the site spacing is greater than 2m but no greater than 5m, the target coordinate matching weight is adjusted to 0.55 to 0.65, and the depth-curvature correlation feature matching weight is adjusted to 0.35 to 0.45. When the site spacing is greater than 5m, the matching confidence of the depth-curvature correlation feature is calculated in real time. If the confidence is not lower than 0.8, the depth-curvature correlation feature matching weight is dynamically increased to 0.5 to 0.6. Based on the dynamically weighted constraints, calculate the optimal registration transformation parameters; Perform registration consistency verification: compare the deviation between the registration result based solely on target coordinate constraints and the dynamically weighted registration result. If the deviation is greater than 0.06 mm, re-optimize the weight allocation and iteratively calculate until the deviation of the registered point cloud is no greater than 0.07 mm.

[0008] Preferably, before employing the statistical filtering algorithm, the point cloud filtering module first performs a preprocessing step based on a multi-view consistency check. This preprocessing is performed according to the following steps: Based on the registered point cloud data from multiple sites, the scanning view parameters corresponding to each scanning site are restored, and a virtual observation view of each site is constructed. For any data point P in the point cloud of any scanning site i Based on the viewing parameters of at least one other scanning station, point P is... i Reprojected onto the two-dimensional observation plane of the station, with point P on this two-dimensional observation plane. i Centered on the reprojection position, a search neighborhood with a preset distance tolerance is defined; Determine whether there are point cloud data points collected by another scanning station within the search neighborhood. If so, determine point P. i Point P is retained after passing the multi-view consistency test; if no corresponding point satisfying the conditions exists on the observation plane of all other scanning stations, then point P is determined to be a valid point. i These are transient noise points and are filtered out.

[0009] Preferably, the volume calculation module includes: A reference plane definition unit is used to fit and generate an initial reference plane based on the target ball center coordinates of at least three non-collinear targets; The planar calibration unit has four theoretical height values ​​stored in advance when the targets are set up. The unit compares the initial reference plane with the theoretical height values ​​and calibrates the initial reference plane by least squares method or spatial rotation transformation to minimize the root mean square error between the theoretical height value of the target ball center coordinates and the height value of the corresponding projection point on the calibrated reference plane. The volume calculation unit is used to calculate the volume of the closed space enclosed between the triangular mesh model and the calibrated reference plane.

[0010] Preferably, an iterative optimization loop is provided between the point cloud filtering module and the triangular mesh modeling module: After the triangulation model is initially constructed, the triangulation modeling module calculates the average distance deviation between each triangular facet and the point cloud data in its surrounding preset neighborhood. The original point cloud data corresponding to triangular facets with an average distance deviation greater than a set threshold are marked as potential outlier point sets, where the set threshold is 0.5-2 times the average spacing of the point cloud. The point cloud filtering module receives a set of potential outlier points and performs secondary fine filtering. The point set is used to remove outlier points using a radius filtering algorithm. The removed point cloud data is sent back to the triangulation modeling module for model reconstruction, forming an iterative optimization until the potential outlier point set is empty or the overall change rate of the model is less than the set value.

[0011] Preferably, the secondary fine filtering employs a hierarchical processing method: Calculate the perpendicular distance from each point in the potential outlier set to its corresponding triangular facet; Points with a vertical distance between 0.5 and 1 times the average spacing of the point cloud are marked as minor anomalies, and their positions are smoothly adjusted and projected onto their respective triangular patches along the normal direction. Points with a vertical distance greater than 1 times the average spacing of the point cloud are marked as severe anomalies and directly removed.

[0012] Preferably, the planar calibration unit is configured to perform a dynamic weighted least squares calibration algorithm: The dynamic weighted least squares calibration algorithm assigns different weights to the theoretical height values ​​of the center coordinates of the four target spheres; The weights are assigned based on the reconstruction quality score of each target in the point cloud data. This score is determined by calculating the spherical fitting residual of the target's point cloud. The smaller the fitting residual, the higher the weight. Based on the weights, a weighted least squares calculation is performed to find the optimal reference plane parameters that minimize the weighted root mean square error.

[0013] Preferably, after completing the initial calibration, the planar calibration unit also performs a redundancy check: The unit sequentially excludes each of the four targets, and uses the center coordinates and theoretical height values ​​of the remaining three targets to refit a reference plane; Calculate the average angle and average elevation deviation between the four newly fitted planes and the initial calibration plane fitted from all four targets; If the average included angle is greater than 0.1° or the average elevation deviation is greater than 1mm, it is determined that there is a gross error or the target is incorrectly placed, and a calibration reliability warning is issued. At the same time, four redundant planes and their deviation values ​​from the initial calibration plane are output. If both the average included angle and elevation deviation are less than the threshold, the initial calibration plane is taken as the final result.

[0014] The present invention has at least the following beneficial effects: First, this invention can acquire high-precision test pit point cloud data through a 3D laser scanner, and combined with the multi-module collaborative processing of the control and processing unit, realize the automated measurement of test pit volume. The reasonable arrangement of the four targets ensures the accuracy of point cloud registration. The overall system operation process is clear, and the measurement results are stable and reliable, which can meet the basic accuracy and efficiency requirements of test pit volume measurement.

[0015] Secondly, by employing an adaptive threshold statistical filtering algorithm, this invention enables the point cloud filtering module to flexibly adjust the filtering parameters according to the overall density of the point cloud. This avoids the problems of incomplete denoising or accidental deletion of valid points that may occur when using a fixed threshold in different point cloud density scenarios, further improving the quality of point cloud data and providing better assurance for the accuracy of subsequent triangulation modeling and volume calculation.

[0016] Third, this invention uses a target-depth-curvature correlation feature dual-constraint dynamic weighting algorithm to perform point cloud registration by combining target coordinates with the pit depth-curvature correlation feature, and dynamically adjusts the weights according to the station spacing. At the same time, consistency verification ensures registration accuracy, effectively solving the problem of insufficient accuracy of single target registration when the station spacing is large, and improving the accuracy and stability of multi-site point cloud registration.

[0017] Fourth, this invention, through the collaborative work of the reference plane definition unit, the plane calibration unit, and the volume calculation unit, first fits the initial reference plane, then calibrates it in conjunction with the theoretical height value of the target, ensuring the accuracy of the reference plane, and finally uses a scientific volume calculation method to obtain the volume of the test pit, effectively reducing the influence of the reference plane error on the volume measurement results and improving the accuracy of the test pit volume calculation.

[0018] Fifth, by setting an iterative optimization loop between the point cloud filtering module and the triangular mesh modeling module, this invention can continuously discover and eliminate potential anomalies in the point cloud, gradually optimize the accuracy of the triangular mesh model, avoid modeling errors caused by incomplete initial filtering, and make the triangular mesh model more closely match the real shape of the test pit's inner surface, thus providing a strong guarantee for the accuracy of subsequent volume calculations.

[0019] Other advantages, objectives and features of the present invention will become apparent in part from the following description, and in part from those skilled in the art through study and practice of the invention. Attached Figure Description

[0020] Figure 1 This is a framework diagram of the test pit volume measurement system based on three-dimensional laser scanning according to the present invention; Figure 2 An iterative optimization loop flowchart is provided between the point cloud filtering module and the triangular network modeling module. Detailed Implementation

[0021] The present invention will now be described in further detail with reference to the accompanying drawings, so that those skilled in the art can implement it based on the description.

[0022] According to one embodiment of the present invention, such as Figure 1 As shown, a test pit volume measurement system based on three-dimensional laser scanning includes: The 3D laser scanner has a scanning distance range of 0.5-150m, which can be flexibly selected according to the actual size of the test pit. For example, when the diameter of the test pit is small, a scanning distance of 0.5-50m can be selected, and when the scale of the test pit is large, a scanning distance of 50-150m can be selected. The point cloud density is 10,000 to 500,000 points per square meter. In practical applications, if there are few details on the surface of the test pit, a density of 10,000-100,000 points per square meter can be selected. If it is necessary to accurately capture details such as depressions and protrusions on the inner surface of the test pit, a density of 100,000-500,000 points per square meter can be selected. The 3D laser scanner needs to be set up on a tripod. An adjustable height aluminum alloy tripod can be selected. The tripod should be placed in a flat and unobstructed location around the test pit to ensure that the scanner can be completely aligned with the scanning area of ​​the test pit and avoid scanning blind spots. The control and processing unit has a built-in point cloud registration module, point cloud filtering module, triangulation modeling module, and volume calculation module. The 3D laser scanner establishes a data connection with the control and processing unit through a wireless transmission module. The wireless transmission module can be a Wi-Fi module or a 4G / 5G transmission module. The wireless transmission module can be integrated inside the 3D laser scanner and the control and processing unit, or connected through an external interface to ensure the real-time and reliable data transmission between the two. The control and processing unit can be placed on the operating table near the test pit for convenient parameter setting and data viewing by the staff. Four targets (target spheres) are arranged symmetrically around the center of the test pit at four positions on the outer edge of the test pit. Commonly available spherical reflective targets can be selected. The main material is a metal sphere with a reflective film on the surface. The metal sphere ensures structural stability, and the reflective film can improve laser reflection efficiency and ensure accurate recognition by the scanner. Each target must be placed on a base of uniform height and without obstruction. The base can be a concrete base or a metal bracket. The base must be fixed to the ground to prevent the target from shifting during the measurement process. During measurement: The three-dimensional laser scanner is set up on a tripod and aligned with the test pit scanning area. The three-dimensional laser scanner acquires point cloud data containing the target and the inner surface of the test pit from at least two different scanning stations. For example, the first scanning station is selected on the north side of the test pit and the second scanning station is selected on the east side of the test pit. The point cloud registration module identifies and matches the spatial coordinates of the target sphere centers of four targets in the point cloud data of different scanning stations, and registers the point cloud data to the same coordinate system. During the identification process, it is necessary to ensure that the targets are clearly visible in the point cloud data. The point cloud filtering module uses a statistical filtering algorithm to denoise the registered point cloud data. The threshold range for judging the number of neighboring points in the statistical filtering algorithm is 5-50. In actual operation, it can be selected according to the point cloud density. For example, when the point cloud density is low, the threshold is set to 5-20, and when the point cloud density is high, the threshold is set to 20-50. The triangular mesh modeling module is used to construct a triangular mesh model from the denoised point cloud data; The volume calculation module calculates the volume of the enclosed space between the triangular mesh model and the initial reference plane, which is generated by fitting the center coordinates of at least three non-collinear target spheres. Using this technical solution, the present invention can acquire high-precision test pit point cloud data using a 3D laser scanner. Combined with multi-module collaborative processing in the control and processing unit, it achieves automated measurement of test pit volume. The reasonable arrangement of the four targets ensures the accuracy of point cloud registration. The overall system operation process is clear, and the measurement results are stable and reliable, meeting the basic accuracy and efficiency requirements for test pit volume measurement.

[0023] According to another embodiment of the present invention, the statistical filtering algorithm of the point cloud filtering module is an adaptive threshold statistical filtering algorithm; The adaptive threshold statistical filtering algorithm first calculates the overall point density of the registered point cloud and then dynamically adjusts the threshold used to determine the number of neighboring points based on this overall point density. The higher the point density, the larger the threshold becomes. The core of this algorithm is to dynamically adjust the threshold for determining the number of neighboring points based on the overall point density of the point cloud data. In practical applications, the overall point density of the registered point cloud needs to be calculated first. The calculation method involves selecting multiple representative regions from the test pit point cloud data, counting the number of points and the area of ​​each region, and then averaging the results to obtain the overall point density. For example, selecting three regions—the bottom of the test pit, the middle of the pit wall, and the top of the pit wall—with each region having an area of ​​1 square meter, and then counting the number of points in each region and taking the average, yields the overall point density. After calculating the overall point density, the threshold for the number of neighboring points is dynamically adjusted based on the point density. If the overall point density is 10,000-100,000 points per square meter, the point cloud distribution is relatively sparse, and the threshold for the number of neighboring points can be set to 5-20 to avoid accidentally deleting valid point cloud data. If the overall point density is 100,000-300,000 points per square meter, the point cloud distribution is at a medium density, and the threshold can be set to 20-35. If the overall point density is 300,000-500,000 points per square meter, the point cloud distribution is relatively dense, and the threshold can be set to 35-50. A standard test pit model with a known volume can be selected, and the point cloud data of the model can be obtained using a 3D laser scanner. Noise removal is performed using both a fixed threshold statistical filtering algorithm and an adaptive threshold statistical filtering algorithm. Then, the differences between the triangular mesh model constructed from the point cloud data processed by the two algorithms and the standard test pit model, as well as the deviation between the calculated volume and the standard volume, are compared. By conducting multiple tests on the processing effects of two algorithms under different point cloud densities, the adaptability and denoising effect of the adaptive threshold statistical filtering algorithm in different point cloud density scenarios were verified. This ensures that the algorithm can dynamically adjust the threshold according to the point cloud density, effectively removing noisy points while retaining valid point cloud data. Using this technical solution, this invention, through the adoption of the adaptive threshold statistical filtering algorithm, enables the point cloud filtering module to flexibly adjust the filtering parameters according to the overall point cloud density. This avoids the problems of incomplete denoising or accidental deletion of valid points that may occur with a fixed threshold in different point cloud density scenarios, further improving the quality of point cloud data and providing better assurance for the accuracy of subsequent triangulation modeling and volume calculation.

[0024] According to another embodiment of the present invention, the point cloud filtering module further integrates a radius filtering algorithm; The execution logic of the point cloud filtering module is as follows: First, a statistical filtering algorithm is used for the first stage of denoising to remove discrete noise points; then, a radius filtering algorithm is used for the second stage of denoising. The search radius of the radius filtering is set to 2-3 times the average point spacing of the point cloud to further remove outlier noise clusters. The key parameter of the radius filtering algorithm is the search radius, which needs to be set to 2-3 times the average point spacing of the point cloud. The average point spacing of the point cloud is calculated based on the point cloud density. If the point cloud density is N points per square meter, the average point spacing is 1 divided by the square root of N. For example, when the point cloud density is 10,000 points per square meter, the average point spacing is 0.01 meters, and the search radius can be set to 0.02-0.03 meters; if the point cloud density is 250,000 points per square meter, the average point spacing is 0.002 meters, and the search radius is set to 0.004-0.006 meters. The radius filtering algorithm searches for other point cloud data points within a set search radius. If a point has no other points within the search radius, it is identified as an outlier noise cluster and further eliminated. This invention employs a two-stage denoising logic: first, an adaptive threshold statistical filtering algorithm is used to eliminate discrete noise points, and then a radius filtering algorithm is used to eliminate outlier noise clusters. This results in a more thorough point cloud filtering effect, effectively reducing the impact of noise points on subsequent triangulation modeling and volume calculation, and further improving the accuracy of test pit volume measurement.

[0025] According to another embodiment of the present invention, the point cloud registration module is configured to execute a target-surface feature dual-constraint dynamic weighting algorithm, the target-surface feature dual-constraint dynamic weighting algorithm comprising the following steps: Based on the spatial coordinates of the target sphere centers of the four targets, the initial registration transformation parameters are calculated. Two to three sets of depth-curvature correlation features were extracted from the point cloud data of each scanning station to the inner surface of the test pit. The depth-curvature correlation features include the continuous concave contour at a specific depth of the pit wall and the slope change zone between the center and the edge of the pit bottom. A site spacing-feature weight mapping model is constructed: registration weights are dynamically allocated based on the spatial distance between adjacent scanning sites. When the site spacing is no greater than 2m, the target coordinate matching weight is set to 0.75 to 0.85, and the depth-curvature correlation feature matching weight is set to 0.15 to 0.25. When the site spacing is greater than 2m but no greater than 5m, the target coordinate matching weight is adjusted to 0.55 to 0.65, and the depth-curvature correlation feature matching weight is adjusted to 0.35 to 0.45. When the site spacing is greater than 5m, the matching confidence of the depth-curvature correlation feature is calculated in real time. If the confidence is not lower than 0.8, the depth-curvature correlation feature matching weight is dynamically increased to 0.5 to 0.6. Based on the dynamically weighted constraints, calculate the optimal registration transformation parameters; Perform registration consistency verification: Compare the deviation between the registration result based solely on target coordinate constraints and the dynamically weighted registration result. If the deviation is greater than 0.06mm, re-optimize the weight allocation and iteratively calculate until the deviation of the registered point cloud is no greater than 0.07mm. Before measurement, the four targets were arranged as required at the four positions on the outer edge of the test pit. After the 3D laser scanner acquired point cloud data from different scanning stations, the point cloud registration module first identified the targets in the point cloud data of each station. By extracting the point cloud data of the target spheres, the spherical fitting algorithm was used to calculate the spatial coordinates of the target sphere center of each target. After the center coordinates of the four targets were determined, the iterative nearest point algorithm was used to calculate the initial registration transformation parameters based on these coordinates. The initial registration transformation parameters include translation vectors and rotation matrices, which are used to initially map the point cloud data of different scanning stations to similar coordinate systems. From the point cloud data of each scanning station, extract 2 to 3 sets of depth-curvature correlation features of the inner surface of the test pit. In practice, the continuous concave contour at a specific depth of the pit wall can be selected as the first set of features, for example, the contour at a depth of 0.5 m. This contour is usually formed by continuous concavities during the excavation of the test pit. The slope abrupt change zone between the center and edge of the pit bottom can be selected as the second set of features. The central area of ​​the pit bottom is relatively flat, and the slope changes significantly when transitioning to the pit wall at the edge area, forming a slope abrupt change zone. If there are other obvious feature structures on the inner surface of the test pit, such as raised edges, they can be used as the third set of features. When extracting these features, the point cloud data range of the feature area can be determined by curvature calculation and depth analysis of the point cloud data, forming a feature point set. By adopting this technical solution, the present invention uses a target-depth-curvature correlation feature dual-constraint dynamic weighting algorithm to perform point cloud registration by combining target coordinates with the pit depth-curvature correlation feature, and dynamically adjusts the weights according to the station spacing. At the same time, consistency verification is used to ensure registration accuracy, which effectively solves the problem of insufficient accuracy of single target registration when the station spacing is large, and improves the accuracy and stability of multi-site point cloud registration.

[0026] According to another embodiment of the present invention, before employing a statistical filtering algorithm, the point cloud filtering module first performs a preprocessing based on a multi-view consistency check, the preprocessing being performed according to the following steps: Based on the registered multi-site point cloud data, the scanning perspective parameters corresponding to each scanning site are restored, and the virtual observation perspective of each site is constructed. The scanning perspective parameters include the scanner's position coordinates, scanning direction angle, lens focal length, etc. These parameters can be obtained from the measurement log of the 3D laser scanner. Then, based on these parameters, the virtual observation perspective of each site is constructed in the software of the control processing unit. The virtual observation perspective can simulate the observation range and angle of each site during actual scanning. For any data point P in the point cloud of any scanning site i Based on the viewpoint parameters of at least one other scanning station, point P is transformed using a spatial coordinate transformation algorithm. i The reprojection is then applied to the two-dimensional observation plane of the site. The reprojection process must consider both intrinsic (e.g., focal length, pixel size) and extrinsic (e.g., position, angle) parameters of the scanner to ensure the accuracy of the reprojection position. On this two-dimensional observation plane, point P is used as the reference point. i Centered on the reprojection position, a search neighborhood with a preset distance tolerance is defined. The preset distance tolerance can be set according to the scanning accuracy of the 3D laser scanner, usually set to 0.01-0.03mm. For example, when the scanner accuracy is 0.02mm, the distance tolerance is set to 0.02mm. Determine whether there are point cloud data points collected by another scanning station within the search neighborhood (the determination process is to compare whether the point cloud data corresponding to the pixels in the neighborhood contains points from other stations). If so, determine point P. i The point is retained after passing the multi-view consistency test, indicating that it is a true point on the surface of the test pit; if there is no corresponding point satisfying the conditions on the observation plane of all other scanning stations, then point P is determined to be a true point. i These are transient noise points, often caused by accidental light reflections or temporary equipment interference during scanning. During preprocessing, this operation can be performed sequentially on each data point in the point cloud from all scanning stations, ensuring that preprocessing covers all point cloud data. By adding a multi-view consistency check preprocessing step before statistical filtering, this invention can filter out transient noise points in advance, reducing the processing burden on subsequent statistical filtering algorithms. It also avoids interference from transient noise points on statistical filtering parameter settings, further improving the overall effect of point cloud filtering and providing a higher-quality point cloud data foundation for subsequent point cloud registration optimization and modeling calculations.

[0027] According to another embodiment of the present invention, the volume calculation module includes: A reference plane definition unit is used to fit and generate an initial reference plane based on the target ball center coordinates of at least three non-collinear targets; The plane calibration unit has four theoretical height values ​​stored in advance when the targets are set up (these theoretical height values ​​are obtained by measuring with a high-precision level when the targets are set up. During the measurement, the level is placed in a horizontal position around the test pit, and the elevation of the top of each target is measured in turn and recorded as the theoretical height value). This unit compares the initial reference plane with the theoretical height values ​​and calibrates the initial reference plane by the least squares method or spatial rotation transformation to minimize the root mean square error between the theoretical height value of the target ball center coordinates and the height value of the corresponding projection point on the calibrated reference plane. The volume calculation unit is used to calculate the volume of the closed space enclosed by the triangular mesh model and the calibrated reference plane. During the calculation, the intersection line between the triangular mesh model and the reference plane is first determined. This intersection line divides the triangular mesh model into portions located above and below the reference plane. Since the test pit is a concave structure, the volume calculation mainly focuses on the portion of the triangular mesh model located below the reference plane. Subsequently, a spatial integration algorithm (such as the tetrahedral volume integration method) is used to divide the closed space into multiple small tetrahedra. The volume of each small tetrahedron is calculated and summed to obtain the total volume of the test pit. During the calculation process, it must be ensured that there are no gaps in the closed space between the triangular mesh model and the reference plane. If gaps exist, the model must be reconstructed from the triangular mesh modeling module. The main function of the reference plane definition unit is to generate the initial reference plane. During operation, the target center coordinates of four targets are first extracted from the point cloud data. At least three non-collinear target center coordinates are selected, such as the target center coordinates in the east, south, and north directions. A plane fitting algorithm (such as the least squares plane fitting algorithm) is used to fit these coordinates to generate the initial reference plane. During the fitting process, it is necessary to ensure that the selected target center coordinates have no obvious deviation. If a target coordinate is abnormal, the target can be excluded and other three non-collinear targets can be selected. The planar calibration unit compares the initial reference plane generated by the reference plane definition unit with the pre-stored theoretical height values. It calculates the difference between the theoretical height value of each target center coordinate and the height value of the corresponding projection point on the initial reference plane. Then, it calibrates the initial reference plane using either the least squares method or spatial rotation transformation. If the least squares method is used, an error function needs to be constructed to minimize the root mean square error of the height differences between all targets. If spatial rotation transformation is used, the spatial angle of the initial reference plane needs to be adjusted until the root mean square error of the height differences between all targets meets the requirements. The calibration process is automatically executed in the software of the control processing unit without manual intervention. Using this technical solution, the present invention, through the collaborative work of the reference plane definition unit, the planar calibration unit, and the volume calculation unit, first fits the initial reference plane, then calibrates it using the theoretical height values ​​of the targets, ensuring the accuracy of the reference plane. Finally, a scientific volume calculation method is used to obtain the test pit volume, effectively reducing the impact of reference plane errors on the volume measurement results and improving the accuracy of the test pit volume calculation.

[0028] According to yet another embodiment of the present invention, such as Figure 2 As shown, an iterative optimization loop is provided between the point cloud filtering module and the triangular network modeling module: After initially constructing the triangulation model, the triangulation modeling module calculates the average distance deviation between each triangular facet and the point cloud data within its preset neighborhood. The range of the preset neighborhood can be set according to the average point spacing of the point cloud, and is usually set to include 3-5 adjacent point cloud data points around the triangular facet. The average distance deviation is calculated by calculating the vertical distance from each point cloud data point in the neighborhood to the triangular facet, and then averaging these vertical distances. (The triangulation modeling module uses the Delaunay triangulation algorithm to construct the triangulation model from the point cloud data after the point cloud filtering module has removed noise. During the construction process, it is necessary to ensure the integrity of the point cloud data to avoid loopholes in the triangulation model due to missing point clouds.) The original point cloud data corresponding to triangular facets with an average distance deviation greater than a set threshold are marked as potential outlier sets. The set threshold is 0.5-2 times the average spacing of the point cloud, which is calculated based on the point cloud density. For example, when the average spacing of the point cloud is 0.01 meters, the set threshold can be set to 0.005-0.02 meters. If the average distance deviation of a certain triangular facet is 0.025 meters, which is greater than the set threshold of 0.02 meters, then the original point cloud data corresponding to that triangular facet (i.e., the three points that make up the triangular facet) is marked as a potential outlier set. The point cloud filtering module receives a set of potential outlier points and performs secondary fine filtering. The point set is used to remove outlier points using a radius filtering algorithm. The removed point cloud data is sent back to the triangulation modeling module for model reconstruction, forming an iterative optimization until the potential outlier set is empty or the overall model change rate is less than the set value (the above steps are repeated, and the average distance deviation of each triangular facet in the new triangulation model is calculated again to determine whether there is a new potential outlier set. If the potential outlier set is empty, it means that the point cloud data quality has met the modeling requirements, and the iterative optimization stops; if there is still a potential outlier set, the second fine filtering and model reconstruction are performed until the potential outlier set is empty or the overall model change rate is less than the set value. The method for calculating the overall model change rate is to compare the volume difference and surface area difference of the triangulation model constructed in the two iterations and take the average value as the change rate. The set value is usually set to 0.1%-0.3%. For example, when the change rate is 0.2%, it is less than the set value of 0.3%, and the iterative optimization stops). By adopting this technical solution, the present invention can continuously discover and eliminate potential anomalies in the point cloud by setting an iterative optimization loop between the point cloud filtering module and the triangular mesh modeling module, gradually optimize the accuracy of the triangular mesh model, avoid modeling errors caused by incomplete initial filtering, and make the triangular mesh model more closely fit the real morphology of the test pit surface, thus providing a strong guarantee for the accuracy of subsequent volume calculation.

[0029] According to another embodiment of the present invention, the secondary fine filtering adopts a hierarchical processing method: Calculate the perpendicular distance from each point in the potential outlier set to its corresponding triangular facet. For any data point in the potential outlier set, calculate the perpendicular distance from the point to the triangular facet using spatial geometry formulas based on the coordinates of the three vertices of the triangular facet. During the calculation, it is necessary to ensure that the vertex coordinates of the triangular facet are accurate. If there is a deviation in the vertex coordinates of the triangular facet, the correct vertex coordinates must be obtained again before the calculation is performed. Points with a vertical distance between 0.5 and 1 times the average point cloud spacing are marked as minor anomalies. These anomalies are then smoothed by projecting their positions onto their respective triangular facets along the normal direction. For example, when the average point cloud spacing is 0.01 m, points with a vertical distance between 0.005 and 0.01 m are considered minor anomalies. The smoothing adjustment method involves projecting the position of a point onto its respective triangular facet along the normal direction. The normal direction is calculated using the cross product of the three vertices of the triangular facet. During the projection process, it must be ensured that the projected position of the point accurately falls within the triangular facet's area. If the projected position exceeds the triangular facet's area, the vertical distance calculation result and the normal direction must be rechecked for accuracy. Points with a vertical distance greater than 1 times the average spacing of the point cloud are marked as severe anomalies and directly removed. For example, when the average spacing of the point cloud is 0.01 m, points with a vertical distance greater than 0.01 m are considered severe anomalies and are directly removed. The removal process is automatically executed in the point cloud filtering module software. After removal, the number and location of severe anomalies need to be recorded for subsequent analysis of the causes of anomalies, such as whether they are caused by strong interference during scanning or target displacement. After the hierarchical processing is completed, the point cloud filtering module sends the processed point cloud data to the triangulation modeling module for reconstruction of the triangulation model, ensuring that the accuracy of the reconstructed model is improved. By adopting this technical solution, this invention performs hierarchical processing on potential anomalies. For minor anomalies, a smooth adjustment method is used to retain point cloud data, avoiding excessive removal that leads to point cloud loss. Severe anomalies are directly removed, ensuring the quality of point cloud data and further optimizing the effect of secondary fine filtering, making the processed point cloud data more suitable for constructing high-precision triangulation models.

[0030] According to another embodiment of the present invention, the planar calibration unit is configured to perform a dynamic weighted least squares calibration algorithm: The dynamic weighted least squares calibration algorithm assigns different weights to the theoretical height values ​​of the center coordinates of the four target spheres; The weights are assigned based on the reconstruction quality score of each target in the point cloud data. This score is determined by calculating the spherical fitting residual of the target's point cloud. The smaller the fitting residual, the higher the weight. Based on the weights, a weighted least squares calculation is performed to find the optimal reference plane parameters that minimize the weighted root mean square error. When calculating the spherical fitting residuals, the point cloud data of each target is first extracted, and the theoretical sphere of the target sphere is fitted using a spherical fitting algorithm. Then, the distance from each target point cloud data point to the theoretical sphere is calculated. The root mean square value of these distances is the spherical fitting residual. The smaller the fitting residual, the closer the target point cloud data is to the actual target sphere shape, the better the reconstruction quality, and the higher the corresponding weight. For example, if the spherical fitting residuals of the four targets are 0.01mm, 0.02mm, 0.015mm, and 0.025mm, respectively, the target with the smallest fitting residual (0.01mm) can be weighted at 0.3, the second smallest at 0.015mm at 0.25, the largest at 0.02mm at 0.25, and the largest at 0.025mm at 0.2. The total weight of the four targets is 1. After weight allocation, weighted least squares calculations are performed based on these weights to construct a weighted error function. This error function is based on the difference between the theoretical height of each target and the height of the projection point on the reference plane, multiplied by the corresponding weight, and then summed. By solving for the minimum value of the error function, the optimal reference plane parameters that minimize the weighted root mean square error are obtained. Using this technical solution, this invention employs a dynamic weighted least squares calibration algorithm to allocate weights according to the target reconstruction quality, making the calibrated reference plane more accurately reflect the true reference surface. This reduces the reference plane error caused by poor reconstruction quality of some targets, further improving the accuracy of the reference plane and providing a more reliable reference for calculating the test pit volume.

[0031] According to another embodiment of the present invention, after completing the initial calibration, the plane calibration unit also performs a redundancy check to ensure the reliability of the reference plane calibration results: The unit sequentially excludes each of the four targets, and uses the center coordinates and theoretical height values ​​of the remaining three targets to refit a reference plane; Calculate the average angle and average elevation deviation between the four newly fitted planes and the initial calibration plane fitted from all four targets; If the average included angle is greater than 0.1° or the average elevation deviation is greater than 1mm, it is determined that there is a gross error or the target is placed incorrectly. A calibration reliability warning is issued, and four redundant planes and their deviation values ​​from the initial calibration plane are output at the same time, including the included angle and elevation deviation of each redundant plane from the initial calibration plane, so that staff can troubleshoot the cause of the problem. If both the average included angle and elevation deviation are less than the threshold, the initial calibration plane is used as the final result (if the average included angle is no greater than 0.1° and the average elevation deviation is no greater than 1mm, the calibration result is considered reliable, and the initial calibration plane is used as the final result for subsequent volume calculations). The first step of redundancy verification is to sequentially exclude each of the four targets: first, exclude the eastern target, using the center coordinates and theoretical height values ​​of the southern, western, and northern targets; second, exclude the southern target, using the center coordinates and theoretical height values ​​of the eastern, western, and northern targets; third, exclude the western target, using the center coordinates and theoretical height values ​​of the eastern, southern, and northern targets; fourth, exclude the northern target, using the center coordinates and theoretical height values ​​of the eastern, southern, and western targets. A new reference plane is then refitted using the least squares plane fitting algorithm for each of these four new planes, resulting in four new reference planes. The average included angle and average elevation deviation between the four newly fitted planes and the initial calibration plane fitted from all four targets are calculated. The average included angle is calculated by calculating the dihedral angle between each newly fitted plane and the initial calibration plane. The dihedral angle is obtained by calculating the dot product of the normal vectors of the two planes. The average included angle is then calculated by averaging the four dihedral angles. The average elevation deviation is calculated by selecting multiple evenly distributed reference points around the test pit, calculating the elevation value of each reference point on the newly fitted plane and the initial calibration plane, and calculating the difference between the two. The average elevation deviation is then calculated by averaging the differences of all reference points. Typically, 10-20 reference points are selected to ensure the representativeness of the calculation results. Using this technical solution, this invention, through redundant verification, can promptly detect gross errors or target placement errors that may exist during the reference plane calibration process, preventing incorrect reference planes from being used for volume calculations. This further ensures the reliability and accuracy of the test pit volume measurement results and reduces the risk of measurement errors caused by calibration errors. By employing this technical solution, the present invention can promptly detect gross errors or target placement errors that may exist during the calibration of the reference plane by performing redundant verification, thereby preventing incorrect reference planes from being used for volume calculations. This further ensures the reliability and accuracy of the test pit volume measurement results and reduces the risk of measurement errors caused by calibration errors.

[0032] The number of devices and processing scale described herein are for simplification of the invention. Applications, modifications, and variations of the present invention's three-dimensional laser scanning-based test pit volume measurement system will be readily apparent to those skilled in the art.

[0033] Although embodiments of the present invention have been disclosed above, they are not limited to the applications listed in the specification and embodiments. They can be applied to various fields suitable for the present invention. For those skilled in the art, other modifications can be easily made. Therefore, without departing from the general concept defined by the claims and their equivalents, the present invention is not limited to the specific details and illustrations shown and described herein.

Claims

1. A test pit volume measurement system based on three-dimensional laser scanning, characterized in that, include: The 3D laser scanner has a scanning distance range of 0.5-150m and a point cloud density of 10,000 to 500,000 points per square meter. The control and processing unit has a built-in point cloud registration module, point cloud filtering module, triangulation modeling module and volume calculation module. The 3D laser scanner establishes a data connection with the control and processing unit through a wireless transmission module. Four targets are arranged symmetrically around the center of the test pit at four locations on the outer edge of the test pit. During measurement: The three-dimensional laser scanner is set up on a tripod and aligned with the test pit scanning area. The three-dimensional laser scanner acquires point cloud data containing the target and the inner surface of the test pit from at least two different scanning stations. The point cloud registration module identifies and matches the spatial coordinates of the target sphere centers of four targets in the point cloud data of different scanning stations, and registers the point cloud data to the same coordinate system. The point cloud filtering module uses a statistical filtering algorithm to denoise the registered point cloud data. The threshold range for determining the number of neighboring points in the statistical filtering algorithm is 5-50. The triangular mesh modeling module is used to construct a triangular mesh model from the denoised point cloud data; The volume calculation module calculates the volume of the enclosed space between the triangular mesh model and the initial reference plane, based on the spatial relationship between the triangular mesh model and the initial reference plane, which is generated by fitting the center coordinates of the target spheres of at least three non-collinear targets. The point cloud registration module is configured to execute a target-surface feature dual-constraint dynamic weighting algorithm, which includes the following steps: Based on the spatial coordinates of the target sphere centers of the four targets, the initial registration transformation parameters are calculated. Two to three sets of depth-curvature correlation features were extracted from the point cloud data of each scanning station to the inner surface of the test pit. The depth-curvature correlation features include the continuous concave contour at a specific depth of the pit wall and the slope change zone between the center and the edge of the pit bottom. A site spacing-feature weight mapping model is constructed: registration weights are dynamically allocated based on the spatial distance between adjacent scanning sites. When the site spacing is no greater than 2m, the target coordinate matching weight is set to 0.75 to 0.85, and the depth-curvature correlation feature matching weight is set to 0.15 to 0.

25. When the site spacing is greater than 2m but no greater than 5m, the target coordinate matching weight is adjusted to 0.55 to 0.65, and the depth-curvature correlation feature matching weight is adjusted to 0.35 to 0.

45. When the site spacing is greater than 5m, the matching confidence of the depth-curvature correlation feature is calculated in real time. If the confidence is not lower than 0.8, the depth-curvature correlation feature matching weight is dynamically increased to 0.5 to 0.

6. Based on the dynamically weighted constraints, the optimal registration transformation parameters are calculated. Perform registration consistency verification: compare the deviation between the registration result based solely on target coordinate constraints and the dynamically weighted registration result. If the deviation is greater than 0.06 mm, re-optimize the weight allocation and iteratively calculate until the deviation of the registered point cloud is no greater than 0.07 mm.

2. The test pit volume measurement system based on three-dimensional laser scanning as described in claim 1, characterized in that, The statistical filtering algorithm of the point cloud filtering module is an adaptive threshold statistical filtering algorithm; The adaptive threshold statistical filtering algorithm first calculates the overall point density of the registered point cloud, and then dynamically adjusts the threshold used to determine the number of neighboring points based on the overall point density. The higher the point density, the larger the threshold is.

3. The test pit volume measurement system based on three-dimensional laser scanning as described in claim 2, characterized in that, The point cloud filtering module also integrates a radius filtering algorithm; The execution logic of the point cloud filtering module is as follows: first, a statistical filtering algorithm is used for the first stage of denoising to remove discrete noise points; then, a radius filtering algorithm is used for the second stage of denoising. The search radius of the radius filtering is set to 2-3 times the average point spacing of the point cloud to further remove outlier noise clusters.

4. The test pit volume measurement system based on three-dimensional laser scanning as described in claim 1, characterized in that, Before employing the statistical filtering algorithm, the point cloud filtering module first performs a preprocessing step based on multi-view consistency verification. This preprocessing is performed according to the following steps: Based on the registered point cloud data from multiple sites, the scanning view parameters corresponding to each scanning site are restored, and a virtual observation view of each site is constructed. For any data point P in any scan station point cloud i , according to the perspective parameter of at least one other scan station, the point P i is re-projected to the two-dimensional observation plane of the station, and in the two-dimensional observation plane, a search neighborhood with a preset distance tolerance is defined with the re-projected position of the point P i as the center. Determine whether there are point cloud data points collected by another scanning station within the search neighborhood. If so, determine point P. i It was retained after passing the multi-perspective consistency test; If there is no corresponding point satisfying the conditions on the observation plane of all other scanning stations, then the decision point P is determined. i These are transient noise points and are filtered out.

5. The test pit volume measurement system based on three-dimensional laser scanning as described in claim 4, characterized in that, The volume calculation module includes: A reference plane definition unit is used to fit and generate an initial reference plane based on the target ball center coordinates of at least three non-collinear targets; The planar calibration unit has four theoretical height values ​​stored in advance when the targets are set up. The unit compares the initial reference plane with the theoretical height values ​​and calibrates the initial reference plane by least squares method or spatial rotation transformation to minimize the root mean square error between the theoretical height value of the target ball center coordinates and the height value of the corresponding projection point on the calibrated reference plane. The volume calculation unit is used to calculate the volume of the closed space enclosed between the triangular mesh model and the calibrated reference plane.

6. The test pit volume measurement system based on three-dimensional laser scanning as described in claim 3, characterized in that, An iterative optimization loop is provided between the point cloud filtering module and the triangular network modeling module: After the triangulation model is initially constructed, the triangulation modeling module calculates the average distance deviation between each triangular facet and the point cloud data in its surrounding preset neighborhood. The original point cloud data corresponding to triangular facets with an average distance deviation greater than a set threshold are marked as potential outlier point sets, where the set threshold is 0.5-2 times the average spacing of the point cloud. The point cloud filtering module receives a set of potential outlier points and performs secondary fine filtering. The point set is used to remove outlier points using a radius filtering algorithm. The removed point cloud data is sent back to the triangulation modeling module for model reconstruction, forming an iterative optimization until the potential outlier point set is empty or the overall change rate of the model is less than the set value.

7. The test pit volume measurement system based on three-dimensional laser scanning as described in claim 6, characterized in that, The secondary fine filtering adopts a hierarchical processing method: Calculate the perpendicular distance from each point in the potential outlier set to its corresponding triangular facet; Points with a vertical distance between 0.5 and 1 times the average spacing of the point cloud are marked as minor anomalies, and their positions are smoothly adjusted and projected onto their respective triangular patches along the normal direction. Points with a vertical distance greater than 1 times the average spacing of the point cloud are marked as severe anomalies and directly removed.

8. The test pit volume measurement system based on three-dimensional laser scanning as described in claim 5, characterized in that, The planar calibration unit is configured to perform a dynamic weighted least squares calibration algorithm: The dynamic weighted least squares calibration algorithm assigns different weights to the theoretical height values ​​of the center coordinates of the four target spheres; The weights are assigned based on the reconstruction quality score of each target in the point cloud data. This score is determined by calculating the spherical fitting residual of the target's point cloud. The smaller the fitting residual, the higher the weight. Based on the weights, a weighted least squares calculation is performed to find the optimal reference plane parameters that minimize the weighted root mean square error.

9. The test pit volume measurement system based on three-dimensional laser scanning as described in claim 8, characterized in that, After completing the initial calibration, the planar calibration unit also performs a redundancy check: The unit sequentially excludes each of the four targets, and uses the center coordinates and theoretical height values ​​of the remaining three targets to refit a reference plane; Calculate the average angle and average elevation deviation between the four newly fitted planes and the initial calibration plane fitted from all four targets; If the average included angle is greater than 0.1° or the average elevation deviation is greater than 1mm, it is determined that there is a gross error or the target is incorrectly placed, and a calibration reliability warning is issued. At the same time, four redundant planes and their deviation values ​​from the initial calibration plane are output. If both the average included angle and elevation deviation are less than the threshold, the initial calibration plane is taken as the final result.

Images