An irregular pit depth measurement method based on point cloud data

Abstract

The application relates to the technical field of point cloud data processing, in particular to an irregular pit depth measuring method based on point cloud data, which comprises the following steps: acquiring point cloud data of a target area; acquiring filtered point cloud data; performing plane fitting on the filtered point cloud data to acquire an initial plane; correcting the initial plane by using a PCA algorithm to acquire a corrected normal vector; fitting a final plane again according to the filtered point cloud data and the corrected normal vector; and measuring the distance from the final plane to obtain the depth of the pit. The application combines advanced point cloud processing technologies, in particular filtering algorithms and fitting optimization algorithms, and realizes accurate measurement of the depth of irregular pits in a complex environment.

Description

Technical Field

[0001] This invention relates to the field of point cloud data processing technology, and more specifically to a method for measuring the depth of irregular pits based on point cloud data. Background Technology

[0002] With the continuous advancement of science and technology, lidar, with its ability to acquire and maintain high-precision and high-resolution three-dimensional information about targets, is playing an increasingly important role in various fields such as topographic mapping, autonomous driving, and environmental monitoring. As an advanced remote sensing technology, lidar measures distance by emitting laser pulses. Its basic principle is to send a laser beam to the target and receive the reflected light signal, measuring the time from laser emission to reception to accurately obtain distance information. Therefore, its ability to acquire high-precision three-dimensional information and accurately measure distances demonstrates significant practical value and scientific exploratory significance in the detection of pit depths.

[0003] Existing technologies disclose methods for vehicle height observation, employing the Kalman filter algorithm. This method acquires the vehicle height in real-time and accurately by establishing a whole-vehicle model and a vehicle height observer, providing accurate vehicle centroid height parameters for vehicle control strategies. However, the effectiveness of the Kalman filter algorithm largely depends on the accuracy of the model used to describe the system. If the whole-vehicle model is inaccurate or fails to reflect the dynamic behavior of the actual vehicle, the algorithm's performance may be affected. Those skilled in the art have also explored the application of LiDAR in 3D point cloud data generation, ground point cloud segmentation, non-ground point cloud clustering, and target point cloud recognition. Addressing the issue of ground point clouds reducing target recognition speed and accuracy, a segmentation method based on cloth simulation filtering and Euclidean clustering is proposed. This effectively reduces redundant point cloud data input and improves the accuracy of target clustering segmentation. Furthermore, a deep learning-based ResPointNet++ network model is proposed, improving recognition speed and solving the problem of recognizing occluded targets. However, the performance of the deep learning model in the algorithm largely depends on the quality and diversity of the training data. Insufficient or unrepresentative training data will lead to insufficient generalization ability of the model in practical applications. Summary of the Invention

[0004] To address the aforementioned technical problems, this invention primarily focuses on the fact that the performance of deep learning models in existing algorithms largely depends on the quality and diversity of training data. Insufficient or unrepresentative training data can lead to insufficient generalization ability of the model in practical applications. To meet the task requirements of accurately measuring the depth of irregular pits in near-site security applications, this invention provides a method for measuring the depth of irregular pits based on point cloud data. By combining advanced point cloud processing techniques, particularly filtering and fitting optimization algorithms, it achieves accurate measurement of the depth of irregular pits in complex environments.

[0005] The first objective of this invention is to provide a method for measuring the depth of irregular potholes based on point cloud data, comprising: Acquire point cloud data for the target area; The point cloud data is filtered sequentially using the Z-axis segmentation filtering algorithm and the density clustering filtering algorithm to obtain the filtered point cloud data. The initial plane is obtained by performing plane fitting on the filtered point cloud data; The initial plane is corrected using the PCA algorithm to obtain the corrected normal vector; The final plane is obtained by fitting the filtered point cloud data and the corrected normal vectors again. The depth of the pit can be obtained by measuring the distance from the final plane.

[0006] Preferably, the initial plane is obtained according to the following steps: Three points are randomly selected from the filtered point cloud to construct a candidate plane; Calculate the vertical distance of other points in the filtered point cloud relative to the candidate plane in the Z-axis direction; Set a distance threshold and classify all points whose vertical distance is less than the threshold as interior points; By analogy, multiple candidate planes and their corresponding number of interior points can be obtained; The candidate plane with the most interior points exceeding a certain threshold and the lowest value in the Z-axis direction is selected as the initial plane.

[0007] Preferably, the corrected normal vector is obtained according to the following steps: Obtain the center point from the interior points of the initial plane; The covariance matrix is ​​obtained by centering the interior points in the initial plane based on the center point. Perform eigenvalue decomposition on the covariance matrix to obtain eigenvalues ​​and corresponding eigenvectors; The eigenvector corresponding to the largest eigenvalue is selected as the corrected normal vector.

[0008] Preferably, the final plane is obtained according to the following steps: Obtain the centroid of the midpoint of the initial plane; Construct the plane equation based on the centroid of the initial plane midpoint and the corrected normal vector; The constants in the plane equation are obtained from the corrected normal vector and interior points; The final plane is obtained based on the constructed plane equation and the constants in the plane equation.

[0009] Preferably, the point cloud data is filtered using a Z-axis segmentation filtering algorithm, including: The point cloud data is segmented along the Z-axis to form a series of equal-height segmentation modules; For each module, calculate the number of point clouds within it; A point cloud threshold is set. If the number of point clouds in the module is lower than the point cloud threshold, the point cloud data is considered to be located on the pit wall and is filtered out; otherwise, it is considered to be point cloud data at the bottom of the pit and is retained.

[0010] Preferably, the point cloud data after the Z-axis segmentation filtering algorithm is filtered using a density clustering filtering algorithm, including: Select a point from the point cloud data as the starting point and calculate the distance to other points; Define a neighborhood radius and mark all points falling within this radius as neighborhood points; When the number of points in a neighborhood reaches a threshold, that point is marked as the core point, and a new cluster is created around it. Then, the points in the neighborhood are recursively added to that cluster. For points that are not core points but are located in the neighborhood of core points, they are also added to the corresponding clusters; After all points have been visited, clusters with fewer point clouds are considered to be depressions or protrusions at the bottom of the pit and should be filtered out; conversely, those with more point clouds are considered to be point cloud data at the bottom of the pit and should be retained.

[0011] Preferably, point cloud data of the target area is collected by radar. After the collection is completed, voxel downsampling is performed to obtain point cloud data of the target area.

[0012] The second objective of this invention is to provide a system for measuring the depth of irregular potholes based on point cloud data, characterized in that it includes: The data acquisition module is used to acquire point cloud data of the target area; The data denoising module is used to filter the point cloud data sequentially using the Z-axis segmentation filtering algorithm and the density clustering filtering algorithm to obtain the filtered point cloud data. The plane acquisition module is used to perform plane fitting on the filtered point cloud data to obtain an initial plane; the initial plane is corrected using the PCA algorithm to obtain the corrected normal vector; and the final plane is obtained by fitting again based on the filtered point cloud data and the corrected normal vector. The depth measurement module is used to measure the distance from the final plane to obtain the depth of the pit.

[0013] The present invention has at least the following beneficial effects: This invention provides a method for measuring the depth of irregular potholes based on point cloud data. This method is specifically designed for accurate measurement of irregular potholes. Existing point cloud distance measurement methods are mainly applicable to specific target objects, but have large measurement errors for large-scale targets such as potholes. This method combines the morphological characteristics of the pothole, performs specialized Z-axis segmentation filtering and RANSAC plane fitting, and corrects the fitting results to achieve accurate pothole measurement.

[0014] The method provided by this invention performs well in complex environments. This invention measures potholes using morphological features, unlike other methods that require target identification, effectively eliminating interference and demonstrating excellent performance in complex environments.

[0015] This invention supports real-time parameter adjustment via an interactive interface. Users can select appropriate parameters based on real-time feedback, enabling dynamic adjustments to point cloud data downsampling, filtering, and other operations. The parameter adjustment buttons are shown below. Figure 2 The options are "Add Downsampling Operation", "Add Filtering Operation", and "Downsampling Radius".

[0016] The algorithm provided by this invention has low time complexity. The entire process is simple and fast, with minimal time consumption in the initial filtering operation. The first fitting does not require high precision, strictly limiting the number of iterations. The second fitting is supported by the corrected normal vector and the result of the first fitting, requiring fewer iterations to meet the requirements. On an AMD Ryzen 7 3.2GHz CPU, 16GB of RAM, and an Ubuntu 20.04 operating system, testing with Visual Studio Code showed a real-time frame rate of 60 frames per second.

[0017] The method provided by this invention offers high measurement accuracy. Repeated depth measurement tests on pits using this method showed a deviation percentage of approximately 1% (effective measurement distance: 0.5m-5m). Experimental results demonstrate that this algorithm is a highly stable and robust method for measuring the depth of irregular pits. Attached Figure Description

[0018] Figure 1 This is a framework diagram of the irregular pit depth measurement method based on point cloud data in an example of the present invention.

[0019] Figure 2 This is an interface diagram of the lidar irregular pit ranging software v1.0. Detailed Implementation

[0020] In order to illustrate the technical means and effects adopted by the present invention to achieve the intended purpose, the following detailed description is provided in conjunction with the embodiments.

[0021] This invention proposes a method for measuring the depth of irregular potholes based on point cloud data. In the pothole depth measurement process, the point cloud data is filtered using a Z-axis segmentation filtering algorithm and a density clustering filtering algorithm. Then, the remaining point cloud data undergoes preliminary RANAC plane fitting to extract the plane's normal vector, which is then corrected using the PCA algorithm. The corrected normal vector is used to fit the plane again to obtain the final plane, i.e., the bottom of the pothole. The depth of the pothole is obtained by measuring the distance to the final fitted plane. The main components are as follows: Z-axis segmentation and filtering algorithm: The point cloud data is divided into several equal-sized modules along the Z-axis. The number of point clouds in each module is counted. Modules with a number below a certain threshold are considered to be point cloud data on the wall of the pit and are filtered out, while modules with a number above the threshold are considered to be point cloud data at the bottom of the pit and are retained.

[0022] Density clustering filtering algorithm: Select unvisited data points, calculate their distances to other points in the dataset, and label points within their neighborhood radius as neighborhoods; if the number of neighborhood points is not less than a threshold, mark them as core points and create a new cluster; recursively add points within the neighborhood to clusters; for non-core points, add the corresponding cluster to the neighborhood of a core point; select the next unvisited data point and repeat the above steps until all points are visited. After clustering, classes with fewer point cloud data points are considered as depressions or bulges on the bottom of pits and will be filtered out.

[0023] RANAC plane fitting: Three points are randomly selected from the point cloud dataset to construct a candidate plane model. The distances of other points relative to the candidate plane in the Z-axis direction are calculated, and points with distances below a preset threshold are classified as interior points. This process is iterated multiple times, and the number of interior points of each candidate plane is counted in each iteration. Planes with more than a certain threshold of interior points are selected, and the plane with the lowest Z-axis position is selected as the final plane model.

[0024] PCA correction: The point cloud data used for fitting the plane is centered, the covariance matrix of the centered point cloud data is calculated, the covariance matrix is ​​decomposed into eigenvalues ​​and corresponding eigenvectors, and the eigenvector corresponding to the largest eigenvalue is selected as the corrected normal vector.

[0025] To achieve the above objectives, see Figure 1 As shown, the present invention provides a method for measuring the depth of irregular pits based on point cloud data, comprising: S1. Obtain point cloud data for the target area; Point cloud data of the target area is collected by radar. After the collection is completed, voxel downsampling is performed to obtain point cloud data of the target area.

[0026] In this embodiment, point cloud data of the target area is acquired using radar. After acquisition, voxel downsampling is performed to reduce the total number of points in the dataset while maintaining the overall structure of the point cloud. The voxel downsampling formula is shown in (1), where... It is downsampled point cloud data. It is a set of points in a voxel grid. These are points from the original point cloud. Furthermore, a simple Gaussian filter is performed to remove potential noise and outliers, preparing the data for further processing.

[0027] (1) in, It is downsampled point cloud data; It is a point cloud dataset in a voxel grid; These are points in the original point cloud.

[0028] It should be noted that this invention supports real-time parameter adjustment via an interactive interface. Users can select appropriate parameters based on real-time feedback, enabling dynamic adjustments to point cloud data downsampling, filtering, and other operations. The parameter adjustment buttons are shown below. Figure 2 The options are "Add Downsampling Operation", "Add Filtering Operation", and "Downsampling Radius".

[0029] S2. The point cloud data is filtered sequentially using the Z-axis segmentation filtering algorithm and the density clustering filtering algorithm to obtain the filtered point cloud data. The filtering operation on the point cloud data using the Z-axis segmentation filtering algorithm includes: The point cloud data is segmented along the Z-axis to form a series of equal-height segmentation modules; For each module, calculate the number of point clouds within it; A point cloud threshold is set. If the number of point clouds in the module is lower than the point cloud threshold, the point cloud data is considered to be located on the pit wall and is filtered out; otherwise, it is considered to be point cloud data at the bottom of the pit and is retained.

[0030] In this embodiment, the point cloud data is segmented along the Z-axis to form a series of equal-height segmentation modules. The Z-axis segmentation formula is shown in (2), where... It is the first The point set in each module It's the entire point cloud data. It is a point coordinate, It is the height of the module. This represents the number of points in the k-th module. For each module, the number of point clouds within it is calculated, and a threshold is set. If the number of point clouds in a module is lower than this threshold, it is considered that this part of the point cloud may be located on the pit wall and is filtered out; otherwise, it is considered that it is point cloud data at the bottom of the pit and should be retained.

[0031] (2) Specifically, a density clustering filtering algorithm is used to filter the point cloud data after the Z-axis segmentation filtering algorithm, including: Select a point from the point cloud data as the starting point and calculate the distance to other points; Define a neighborhood radius and mark all points falling within this radius as neighborhood points; When the number of points in a neighborhood reaches a threshold, that point is marked as the core point, and a new cluster is created around it. Then, the points in the neighborhood are recursively added to that cluster. For points that are not core points but are located in the neighborhood of core points, they are also added to the corresponding clusters; After all points have been visited, clusters with fewer point clouds are considered to be depressions or protrusions at the bottom of the pit and should be filtered out; conversely, those with more point clouds are considered to be point cloud data at the bottom of the pit and should be retained.

[0032] In this embodiment, a point is selected from the unlabeled point cloud data as the starting point, and its distance to other points in the dataset is calculated; a neighborhood radius is defined, and all points falling within this radius are marked as neighborhood points; if the number of points in a neighborhood reaches a preset threshold, this point is marked as a core point, and a new cluster is created around it, and then the points in the neighborhood are recursively added to the cluster; for those points that are not core points but are located in the neighborhood of a core point, they are also added to the corresponding cluster; this process is repeated until all points have been visited. After clustering, those clusters with fewer point clouds are considered to be the depressions or protrusions at the bottom of the pit and should be filtered out. The key formulas in the density clustering process are shown in (3) and (4), where, It is a point Let S be the set of neighborhood points, and S be the set of core points. It is the neighborhood radius. It is the minimum number of neighborhood points required to define the core point. It is a point and The Euclidean distance between them.

[0033] (3) (4) S3. Perform plane fitting on the filtered point cloud data to obtain the initial plane; The initial plane is obtained according to the following steps: Three points are randomly selected from the filtered point cloud to construct a candidate plane; Calculate the vertical distance of other points in the filtered point cloud relative to the candidate plane in the Z-axis direction; Set a distance threshold and classify all points whose vertical distance is less than the threshold as interior points; By analogy, multiple candidate planes and their corresponding number of interior points can be obtained; The candidate plane with the most interior points exceeding a certain threshold and the lowest value in the Z-axis direction is selected as the initial plane.

[0034] In this embodiment, three points are randomly selected from the remaining point cloud dataset to construct a candidate planar model. Then, the vertical distances of other points in the dataset relative to this candidate plane along the Z-axis are calculated. A distance threshold is set, and all points with a vertical distance less than this threshold are classified as inliers. This operation is repeated multiple times, counting the number of inliers in the candidate plane each time. Among all candidate planes, the plane with the highest number of inliers (exceeding the specific threshold) and the lowest Z-axis position is selected as the initial plane.

[0035] S4. Correct the initial plane using the PCA algorithm to obtain the corrected normal vector; The corrected normal vector is obtained according to the following steps: Obtain the center point from the interior points of the initial plane; The covariance matrix is ​​obtained by centering the interior points in the initial plane based on the center point. Perform eigenvalue decomposition on the covariance matrix to obtain eigenvalues ​​and corresponding eigenvectors; The eigenvector corresponding to the largest eigenvalue is selected as the corrected normal vector.

[0036] In this embodiment, the center point is calculated using formula (5) from the dataset of interior points in the initial plane obtained in S3. Then, the covariance matrix of the centered data is calculated using formula (6), where It is the first in the interior point data One point, It is the covariance matrix. Next, we examine the covariance matrix... Eigenvalues ​​are obtained by performing eigenvalue decomposition using formula (7). and the corresponding feature vectors From these eigenvectors, the eigenvector corresponding to the largest eigenvalue is selected as the corrected normal vector.

[0037] (5) (6) (7) S5. The final plane is obtained by fitting the filtered point cloud data and the corrected normal vector again. The final plane is obtained according to the following steps: Obtain the centroid of the midpoint of the initial plane; Construct the plane equation based on the centroid of the initial plane midpoint and the corrected normal vector; The constants in the plane equation are obtained from the corrected normal vector and interior points; The final plane is obtained based on the constructed plane equation and the constants in the plane equation.

[0038] In this embodiment, a new plane equation (8) is constructed using the centroids of the interior points in the initial plane obtained in S3 and the normal vectors obtained through PCA correction in S4, where A, B, and C are the components of the plane normal vector, and D is a constant of the plane equation. Specifically, the plane parameters are optimized using the least squares method based on the interior points of the point cloud data in the initial plane, i.e., by minimizing formula (9) to obtain the final plane model, where N is the number of the latest interior points. Let A, B, and C be the coordinates of the new interior point, and J be the objective function. A, B, and C have already been corrected by PCA in S4. We only need to update the constant D using formula (10) to obtain the new plane equation. Finally, we can calculate the Z-axis distance from the final plane to the origin.

[0039] (8) (9) (10) S6. The depth of the pit can be obtained by measuring the distance from the origin of the coordinate system to the final plane.

[0040] In summary, this invention provides a method for measuring the depth of irregular pits based on point cloud data, enabling accurate measurement of the depth of irregular pits. During the filtering process, the data is divided along the Z-axis and subjected to density clustering to eliminate terrain interference such as pit walls, depressions, and convex / sub-protrusions. When fitting the pit bottom plane, the PCA algorithm is used to correct the randomly fitted plane and achieve refitting, thus achieving accurate fitting of the pit bottom plane.

[0041] This invention provides a system for measuring the depth of irregular potholes based on point cloud data, comprising: The data acquisition module is used to acquire point cloud data of the target area; The data denoising module is used to filter the point cloud data sequentially using the Z-axis segmentation filtering algorithm and the density clustering filtering algorithm to obtain the filtered point cloud data. The plane acquisition module is used to perform plane fitting on the filtered point cloud data to obtain an initial plane; the initial plane is corrected using the PCA algorithm to obtain the corrected normal vector; and the final plane is obtained by fitting again based on the filtered point cloud data and the corrected normal vector. The depth measurement module is used to measure the distance from the final plane to obtain the depth of the pit.

[0042] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for measuring the depth of irregular potholes based on point cloud data, characterized in that, include: Acquire point cloud data for the target area; The point cloud data is filtered sequentially using the Z-axis segmentation filtering algorithm and the density clustering filtering algorithm to obtain the filtered point cloud data. The initial plane is obtained by performing plane fitting on the filtered point cloud data; The initial plane is corrected using the PCA algorithm to obtain the corrected normal vector; The final plane is obtained by fitting the filtered point cloud data and the corrected normal vectors again. The depth of the pit can be obtained by measuring the distance from the origin of the coordinate system to the final plane; The initial plane is obtained according to the following steps: Three points are randomly selected from the filtered point cloud to construct a candidate plane; Calculate the vertical distance of other points in the filtered point cloud relative to the candidate plane in the Z-axis direction; Set a distance threshold and classify all points whose vertical distance is less than the threshold as interior points; By analogy, multiple candidate planes and their corresponding number of interior points can be obtained; The candidate plane with the most interior points exceeding a certain threshold and the lowest value in the Z-axis direction is selected as the initial plane. The corrected normal vector is obtained according to the following steps: Obtain the center point from the interior points of the initial plane; The covariance matrix is ​​obtained by centering the interior points in the initial plane based on the center point. Perform eigenvalue decomposition on the covariance matrix to obtain eigenvalues ​​and corresponding eigenvectors; Select the eigenvector corresponding to the largest eigenvalue as the corrected normal vector; The point cloud data is filtered using a Z-axis segmentation filtering algorithm, including: The point cloud data is segmented along the Z-axis to form a series of equal-height segmentation modules; For each module, calculate the number of point clouds within it; A point cloud threshold is set. If the number of point clouds in the module is lower than the point cloud threshold, the point cloud data is considered to be located on the pit wall and is filtered out; otherwise, it is considered to be point cloud data at the bottom of the pit and is retained.

2. The method for measuring the depth of irregular pits based on point cloud data according to claim 1, characterized in that, The final plane is obtained according to the following steps: Obtain the centroid of the midpoint of the initial plane; Construct the plane equation based on the centroid of the initial plane midpoint and the corrected normal vector; The constants in the plane equation are obtained from the corrected normal vector and interior points; The final plane is obtained based on the constructed plane equation and the constants in the plane equation.

3. The method for measuring the depth of irregular pits based on point cloud data according to claim 1, characterized in that, The point cloud data after the Z-axis segmentation filtering algorithm is filtered using a density clustering filtering algorithm, including: Select a point from the point cloud data as the starting point and calculate the distance to other points; Define a neighborhood radius and mark all points falling within this radius as neighborhood points; When the number of points in a neighborhood reaches a threshold, that point is marked as the core point, and a new cluster is created around it. Then, the points in the neighborhood are recursively added to that cluster. For points that are not core points but are located in the neighborhood of core points, they are also added to the corresponding clusters; After all points have been visited, clusters with fewer point clouds are considered to be depressions or protrusions at the bottom of the pit and should be filtered out; conversely, those with more point clouds are considered to be point cloud data at the bottom of the pit and should be retained.

4. The method for measuring the depth of irregular pits based on point cloud data according to claim 1, characterized in that, Point cloud data of the target area is collected by radar. After the collection is completed, voxel downsampling is performed to obtain point cloud data of the target area.

5. A system for measuring the depth of irregular potholes based on point cloud data as described in claim 1, characterized in that, include: The data acquisition module is used to acquire point cloud data of the target area; The data denoising module is used to filter the point cloud data sequentially using the Z-axis segmentation filtering algorithm and the density clustering filtering algorithm to obtain the filtered point cloud data. The plane acquisition module is used to perform plane fitting on the filtered point cloud data to obtain the initial plane; The initial plane is corrected using the PCA algorithm to obtain the corrected normal vector; The final plane is obtained by fitting the filtered point cloud data and the corrected normal vectors again. The depth measurement module is used to measure the distance from the final plane to obtain the depth of the pit.

Images