UAV Laser Point Cloud Precision Self-Evaluation Method Based on the Data Itself

Abstract

The present invention relates to the technical field of point cloud data processing and accuracy evaluation for drone laser scanning, and particularly relates to a method for self-evaluating the accuracy of drone laser point clouds based on the data itself. The evaluation process of this method includes the following steps: (1) input the point cloud, (2) perform radius neighborhood search for each point, (3) calculate the normal vector for each point, (4) perform cylindrical neighborhood search for each point, (5) estimate the accuracy of the point cloud for each point; for each point in the input point cloud, the methods in steps (1) to (5) are used to evaluate its accuracy, and the point-by-point accuracy evaluation result of the input point cloud is obtained. The method for evaluating the accuracy of drone laser point clouds of the present invention does not require external reference data, can complete the evaluation of the accuracy of drone laser point clouds automatically and point by point, and can truly reflect the accuracy of drone laser point clouds.

Description

Technical Field

[0001] The present invention relates to the technical field of point cloud data processing and accuracy evaluation of UAV laser scanning, and particularly relates to a method for self-evaluating the accuracy of UAV laser point cloud based on the data itself. Background Technique

[0002] The UAV laser scanning system integrates sensors such as GNSS, INS, camera, and laser scanner on the UAV platform, and can efficiently obtain the three-dimensional space and attribute information of ground objects during the flight of the UAV. The UAV laser scanning system has been widely used in many industries and fields such as surveying and mapping, forestry, and water conservancy, and is one of the most important geospatial information acquisition technologies at present.

[0003] Sensors such as GNSS, INS, and laser scanner in the UAV laser scanning system all have their own measurement accuracies. At the same time, the accuracy of the UAV laser point cloud is also affected by the scanning incident angle, ground object attributes, and ground object surface conditions. Therefore, the accuracy of the laser point cloud obtained by using the UAV laser scanning system is usually difficult to be consistent with the theoretical prediction. In applications, accurate UAV laser point cloud accuracy indicators are usually required, and different applications usually require three-dimensional laser point clouds with different accuracies. Therefore, it is essential to determine / evaluate the accuracy of the UAV laser point cloud.

[0004] The existing methods for evaluating the accuracy of UAV laser point cloud include the following several:

[0005] (1) The method for evaluating the accuracy of UAV laser point cloud based on theoretical model prediction. This method uses the positioning equation of the UAV laser scanning system, combines the theoretical positioning accuracies of sensors such as GNSS, INS, and laser scanner, and theoretically estimates the accuracy of the obtained laser point cloud through the error propagation law. For example, May and Toth proposed "Point positioning accuracy of airborne LiDAR systems: A rigorous analysis" in 2007. Considering the main potential error sources in the airborne laser scanning system, a rigorous theoretical positioning accuracy model of the airborne laser scanning system was established according to the error propagation law to evaluate the theoretical positioning accuracy that the airborne laser scanning system can achieve. Zou Shuangchao et al. proposed "Research on the evaluation model of point cloud position accuracy considering the beam incident angle" in 2015. On the basis of considering the errors of each sensor and taking into account the beam incident angle, a theoretical model for evaluating the point cloud positioning accuracy was established.

[0006] (2) UAV laser point cloud accuracy assessment methods based on external reference (control points, target points, orthophotos) data. Such methods use data such as external control points, target points, and orthophotos as external references, and evaluate the accuracy of UAV laser point clouds by comparing the coordinates of homologous points between the UAV laser point clouds and the external reference data. For example, Zhang Yue proposed the "Research on Key Technologies for Assessing the Accuracy of Airborne LiDAR Based on RANSAC-TLS" in 2020, and evaluated the measurement accuracy of ground 3D laser point clouds by establishing a calibration field and arranging targets in the calibration field. For example, Luo Shengliang proposed the "Research on the Content and Method of Quality Inspection of Airborne LiDAR Point Cloud Data" in 2019, and used network RTK to measure plane point coordinates in the field to evaluate the accuracy of airborne laser point clouds. Gong Yan proposed the "Method for Quality Inspection of Airborne LiDAR Point Cloud Data" in 2020, and evaluated the accuracy of airborne laser point clouds by comparing the characteristic points such as the corner points of buildings in the point cloud with the actual coordinates of the building corner points. Tao Pengjie et al. proposed the "Research on Evaluating the Accuracy of Airborne LiDAR Point Clouds Using High-Precision DLG" in 2019, and used high-precision digital line graphic (DLG) as a geometric reference to evaluate the accuracy of airborne laser point clouds.

[0007] (3) Evaluate the accuracy of the point cloud by fitting the planes in the point cloud and based on the residuals. For example, Sun Deyong et al. proposed the "Test and Analysis of the Plane Fitting Accuracy of Ground 3D Laser Scanning Point Clouds" in 2021, fitted the plane data in the data obtained by using a ground 3D laser scanner, and established the relationship between factors such as color, material, distance, incident angle, and roughness and the scanning accuracy through the fitting accuracy. Tian Maoyi et al. proposed the "Evaluation of the Geolocation Accuracy of Domestic Airborne Dual-Frequency Lidar" in 2018, and evaluated the positioning accuracy of domestic dual-frequency lidar by fitting planes.

[0008] The UAV laser point cloud accuracy assessment method based on theoretical model prediction can estimate the theoretical positioning accuracy of a specific UAV laser scanning system, but this type of method only considers the theoretical positioning accuracy of each sensor, and does not take into account the influence of factors such as GNSS signal displacement and UAV vibration caused by weather conditions on the accuracy of UAV laser point clouds. Therefore, the positioning accuracy evaluated by this type of method is only used as a reference for the positioning accuracy of a certain model of UAV laser scanning system, and cannot be used as the accuracy index of a specific flight of UAV laser point clouds.

[0009] The UAV laser point cloud accuracy assessment methods based on external reference data such as control points, target points, orthophotos, and digital line graphics can stably and reliably evaluate the accuracy of UAV laser point clouds, and are currently the most commonly used UAV laser point cloud accuracy assessment methods. However, this type of method requires measuring control points in the field, and even arranging target points. Even when using orthophotos, digital line graphics, etc. to obtain reference data, it is necessary to manually extract control points from these data, resulting in a large workload, low efficiency, and difficulty in automating the process.

[0010] Automatically evaluating the accuracy of a point cloud by fitting a plane in the point cloud and evaluating according to the residuals can usually be completed automatically. However, not all UAV lidar point clouds contain planar ground objects. For scenarios without planar ground objects, this type of method cannot be used. Even if there are planar ground objects, most of the time, the distribution of planar ground objects is only limited to one or a few places, which cannot meet the principle of uniform distribution in accuracy evaluation and is difficult to comprehensively evaluate the accuracy of the entire UAV lidar point cloud. Summary of the Invention

[0011] The present invention aims to provide a method for evaluating the accuracy of a UAV lidar point cloud that can be fully automated, evaluate the accuracy of the UAV lidar point cloud point by point without external reference data, and can truly reflect the accuracy of the UAV lidar point cloud.

[0012] The evaluation process of the method for self-evaluating the accuracy of a UAV lidar point cloud based on the data itself proposed by the present invention includes the following steps:

[0013] (1) Input point cloud

[0014] The input point cloud is UAV lidar point cloud data containing at least x, y, and z coordinates.

[0015] (2) Point-by-point radius neighborhood search

[0016] An octree structure is established based on the input point cloud data. For each point P(x, y, z) in the input point cloud, the neighborhood points within the sphere with P(x, y, z) as the center and radius r are obtained based on the octree structure and denoted as N(P).

[0017] (3) Point-by-point calculation of the normal vector

[0018] To calculate the normal vector of point P(x, y, z), first calculate the geometric center coordinates of point P and its neighborhood points N(P) within the sphere with radius r through the following formula:

[0019]

[0020] In the above formula, k is the number of neighborhood points N(P) within the sphere with radius r of point P, X i is the coordinate of point P and its neighborhood points N(P) within the sphere with radius r, is the geometric center coordinate.

[0021] After obtaining the coordinates of point P and its neighborhood points N(P) within the sphere with radius r and the geometric center coordinates, calculate the 3×3 covariance matrix S according to the following formula:

[0022]

[0023] Perform singular value decomposition (SVD) on the 3×3 covariance matrix S to obtain its three eigenvalues λ1, λ2, λ3 (λ1≥λ2≥λ3≥0) and their corresponding eigenvectors e1, e2, e3, where e3 is the normal vector of point P, denoted as n.

[0024] (4) Point-by-point cylindrical neighborhood search

[0025] Construct a cylinder with point P as the center and the line passing through point P and parallel to the normal vector n as the axis. The radius of the cylinder is R and the height is H. Search for points within the cylinder range to obtain the cylindrical neighborhood points of point P, denoted as C(P).

[0026] During the search process of C(P), an octree structure is used to accelerate the search process. Starting from the root node of the octree, a depth-first search method is used for the search. If a node does not intersect with the cylinder, there is no need to continue searching its child nodes. If a node intersects with the cylinder, continue to search its child nodes until the leaf nodes. For the leaf nodes that intersect with the cylinder, substitute the x, y, z coordinates of the points into the method described by the following formula to calculate point by point whether the point is inside the cylinder:

[0027]

[0028] In the above formula, q is the point to be tested for whether it is within the cylinder range, p1 is the center point of the bottom surface of the cylinder, p2 is the center point of the top surface of the cylinder, and R is the radius of the cylinder.

[0029] (5) Point-by-point estimation of the accuracy of the point cloud

[0030] Project the cylindrical neighborhood points C(P) of point P onto the central axis of the cylinder.

[0031] After projection, with the projection p of point P as the coordinate origin and the distance from each projection point to point p as d, the mean value of the distances from each projection point to point p can be calculated:

[0032]

[0033] In the above formula, N is the number of points in the cylindrical neighborhood points C(P) of point P, d i is the distance from the i-th point in C(P) to the projection p of point P after being projected onto the central axis of the cylinder, and m is the mean value of the projection distances.

[0034] Then, evaluate the accuracy of point P through the following formula:

[0035]

[0036] For each point in the input point cloud, use the methods in steps (1) to (5) to evaluate its accuracy and obtain the point-by-point accuracy evaluation result of the input point cloud.

[0037] (6) Output the point cloud with precision. The point cloud can be visualized according to the size of the precision value, intuitively showing the precision of each part of the point cloud, and also facilitating the quality inspection of the UAV laser point cloud. For the point cloud that meets the precision standard, it can be directly used for subsequent processing and production. For the point cloud that does not meet the precision standard, the precision of the point cloud can be improved by methods such as control point correction to meet the precision requirements.

[0038] Compared with the prior art, the present invention has the following advantages:

[0039] (1) Compared with the method for evaluating the precision of UAV laser point cloud based on theoretical model prediction, the method of the present invention can evaluate the actual precision of specific data. The evaluation result not only includes the influence of measurement errors of sensors such as GNSS, INS, and laser scanner, but also takes into account the influence of factors such as GNSS signal loss of lock and scanning geometry.

[0040] (2) Compared with the method for evaluating the precision of UAV laser point cloud based on external reference data such as control points, target points, and orthophotos, the method of the present invention does not require external control data, has high efficiency, and can be fully automated.

[0041] (3) Compared with the method of evaluating the precision of the point cloud by fitting the plane in the point cloud and evaluating the precision according to the residuals, the method of the present invention has lower requirements for scene features and does not require flat features.

[0042] (4) The method of the present invention can evaluate the precision of the UAV laser point cloud point by point. After visualizing according to the precision, it can intuitively display the precision of each part of the point cloud. For the point cloud that meets the precision requirements, it can be directly used for subsequent production processing. For the point cloud that does not meet the precision requirements, methods such as control point correction can be used to improve the precision of the point cloud. The present invention provides a simple and external data-free solution for the quality inspection of UAV laser point cloud. Description of the Drawings

[0043] Figure 1 It is the evaluation flow chart of the UAV laser point cloud precision self-evaluation method of the present invention.

[0044] Figure 2 It is the schematic diagram of the cylindrical neighborhood search of the UAV laser point cloud in the embodiment of the present invention.

[0045] Figure 3 It is the effect diagram of projecting the cylindrical neighborhood points onto the cylindrical central axis in the embodiment of the present invention.

[0046] Figure 4 It is the effect after visualizing according to the precision after using the method of the present invention for precision evaluation.

[0047] Figure 5 For Figure 4Cross-sectional thickness effect diagram at the black square Detailed implementation mode

[0048] In order to more clearly understand the above-mentioned objects, features and advantages of the present invention, the present invention will be further described in detail below in conjunction with the accompanying drawings and specific implementation modes. Many specific details are set forth in the following description in order to fully understand the present invention. However, the present invention can also be implemented in other ways different from those described herein. Therefore, the present invention is not limited by the limitations of the specific embodiments disclosed below.

[0049] As Figure 1 shown in the UAV laser point cloud accuracy self-evaluation method based on the data itself, its evaluation process includes the following steps:

[0050] (1) Input point cloud

[0051] The input point cloud is UAV laser point cloud data (i.e., the result data of UAV laser scanning) that at least includes x, y, and z coordinates.

[0052] (2) Point-by-point radius neighborhood search

[0053] An octree structure is established according to the input point cloud data. For each point P(x, y, z) in the input point cloud, neighborhood points within the sphere range with P(x, y, z) as the center of the sphere and a radius of r (r can be set to 0.1 - 0.2 meters, and is set to 0.1 meter in this embodiment) are obtained based on the octree structure, denoted as N(P).

[0054] (3) Point-by-point normal vector calculation

[0055] Calculate the normal vector of point P(x, y, z). First, calculate the geometric center coordinates of point P and its neighborhood points N(P) within the sphere range with a radius of r through the following formula:

[0056]

[0057] In the above formula, k is the number of neighborhood points N(P) within the sphere range with a radius of r for point P, X i is the coordinate of point P and its neighborhood points N(P) within the sphere range with a radius of r, is the geometric center coordinate.

[0058] After obtaining the coordinates and geometric center coordinates of point P and its neighborhood points N(P) within the sphere range with a radius of r, calculate the 3×3 covariance matrix S according to the following formula:

[0059]

[0060] Perform singular value decomposition (SVD) on the 3×3 covariance matrix S to obtain its three eigenvalues λ1, λ2, λ3 (λ1≥λ2≥λ3≥0) and their corresponding eigenvectors e1, e2, e3, where e3 is the normal vector of point P, denoted as n.

[0061] (4) Point-by-point cylindrical neighborhood search

[0062] Construct a cylinder with point P as the center and the line passing through point P and parallel to the normal vector n as the axis. The radius of the cylinder is R (set to 0.1 m in this embodiment) and the height is H (set to 0.4 m in this embodiment). Search for points within the cylinder range to obtain the cylindrical neighborhood points of point P, denoted as C(P), as Figure 2 shown.

[0063] During the search for C(P), an octree structure is used to accelerate the search process. Starting from the root node of the octree, a depth-first search method is used for the search. If a node does not intersect with the cylinder, there is no need to continue searching its child nodes. If a node intersects with the cylinder, continue to search its child nodes until the leaf nodes. For the leaf nodes that intersect with the cylinder, use the method described by the following formula to judge point-by-point whether the point is inside the cylinder:

[0064]

[0065] In the above formula, q is the point to be tested whether it is within the cylinder range, p1 is the center point of the bottom surface of the cylinder, p2 is the center point of the top surface of the cylinder, and R is the radius of the cylinder.

[0066] (5) Point-by-point estimation of the accuracy of the point cloud

[0067] Project the cylindrical neighborhood points C(P) of point P onto the central axis of the cylinder, as Figure 3 shown.

[0068] After projection, with the projection p of point P as the coordinate origin and the distance from each projection point to point p as d, the mean value of the distances from each projection point to point p can be calculated:

[0069]

[0070] In the above formula, N is the number of points in the cylindrical neighborhood points C(P) of point P, d i is the distance from the i-th point in C(P) after projection onto the central axis of the cylinder to the projection p of point P, and m is the mean value of the projection distances.

[0071] The present invention evaluates the accuracy of point P through the following formula:

[0072]

[0073] For each point in the input point cloud, the method in steps (1) to (5) is used to evaluate its accuracy to obtain the point-by-point accuracy evaluation result of the input point cloud.

[0074] (6) Output point cloud with accuracy

[0075] The point cloud can be visualized according to the size of the precision value, intuitively showing the precision of each point cloud, which is convenient for quality inspection of UAV laser point cloud; point clouds that meet the precision standards can be directly used for subsequent processing and production, and point clouds that do not meet the precision standards can be improved through control point correction and other methods to meet the precision requirements.

[0076] The UAV laser point cloud accuracy self-assessment method of the present invention does not require external reference data, eliminates the process of obtaining field control points and selecting indoor points, has high assessment efficiency, is fully automated, and can assess the UAV laser point cloud accuracy point by point. Compared with the method of using control point sampling assessment, it can effectively improve the accuracy of the assessment results. Figure 4 This is a diagram showing the effect of using the method of the present invention to evaluate the accuracy and visualize the area according to the accuracy. It can be intuitively seen which areas have high accuracy and which areas have low accuracy. Figure 5 for Figure 4 The effect diagram of the cross-section thickness in the black box shows that the manually measured cross-section thickness of the point cloud is 11.7 cm. The accuracy of the automatic calculation using the method of the present invention is 11.2 cm (from Figure 4 It can be qualitatively seen that the accuracy is between 10.7cm and 12.9cm), which is comparable to the manually measured value, indicating that the accuracy evaluation method of the present invention can effectively measure the accuracy of the UAV laser point cloud.

Claims

1. An accuracy self - evaluation method for UAV laser point clouds based on the data itself, characterized in that, The evaluation process includes the following steps: (1) The input point cloud is unmanned aerial vehicle (UAV) laser point cloud data that includes at least x, y, and z coordinates; (2) Conduct a radius neighborhood search for each point. An octree structure is established for the input point cloud. For each point P(x, y, z) in the input point cloud, based on the octree structure, obtain the neighborhood points within the sphere with P(x, y, z) as the center and radius r, denoted as N(P); (3) Calculate the normal vector point by point, calculate the normal vector of point P(x, y, z), and obtain the coordinates X of the neighborhood points N(P) within the sphere with point P and radius r i , and then calculate the geometric center coordinates of the neighborhood points N(P) within the sphere with point P and radius r According to the coordinates X of the neighborhood points N(P) within the range of the sphere with point P and radius r i , the geometric center coordinates Calculate the 3×3 covariance matrix S, perform SVD decomposition on the 3×3 covariance matrix S to obtain its three eigenvalues, and sort the three eigenvalues in descending order, denoted as λ1, λ2, and λ3 respectively; the eigenvectors corresponding to λ1, λ2, and λ3 are denoted as e1, e2, and e3, where e3 is the normal vector of point P, denoted as n; (4) Conduct a cylinder neighborhood search for each point. With point P as the center, construct a cylinder with the line passing through point P and parallel to the normal vector n as the axis. The radius of the cylinder is R and the height is H. Search for the points within the cylinder range to obtain the cylinder neighborhood points of point P, denoted as C(P); During the search process of C(P), use the octree structure to accelerate the search process. Starting from the root node of the octree, use the depth-first search method for search. If the node does not intersect with the cylinder, there is no need to continue searching the child nodes. If the node intersects with the cylinder, continue to search the child nodes until the leaf nodes. For the leaf nodes that intersect with the cylinder, it is necessary to determine whether the point is inside the cylinder; (5) Estimate the accuracy of the point cloud for each point. Project the cylinder neighborhood points C(P) of point P onto the central axis of the cylinder. With the projection p of point P as the coordinate origin, and the distance from each projection point to point p as d, the mean value m of the distances from each projection point to point p can be calculated, and then evaluate the accuracy of point P according to the mean value m of the distances from each projection point to point p; For each point in the input point cloud, use the methods in steps (1) to (5) to evaluate its accuracy, and obtain the point-by-point accuracy evaluation result of the input point cloud.

2. The method for self-evaluating the accuracy of the UAV laser point cloud based on the data itself according to claim 1, characterized in that, After obtaining the point-by-point accuracy evaluation result of the input point cloud, output the point cloud with accuracy, visualize the point cloud according to the magnitude of the accuracy value, intuitively display the accuracy of each part of the point cloud, and also facilitate the quality inspection of the UAV laser point cloud.

3. The method for self-evaluating the accuracy of the UAV laser point cloud based on the data itself according to claim 1, wherein In step (3), the geometric center coordinates of the neighborhood point N(P) within the sphere with the point P and radius r are as follows: The calculation formula is as follows: In the above formula, k is the number of neighborhood points N(P) within the sphere of radius r centered at point P, and X i is the coordinate of point P and its neighborhood points N(P) within the sphere of radius r, which is the coordinate of the geometric center; The calculation formula of the covariance matrix S is as follows:

4. The method for self-evaluating the accuracy of the UAV laser point cloud based on the data itself according to claim 1, characterized in that In step (4), use the method described by the following formula to determine point-by-point whether the leaf nodes that intersect with the cylinder are inside the cylinder: In the above formula, q is the point to be tested whether it is within the cylinder range, p1 is the center point of the bottom surface of the cylinder, p2 is the center point of the top surface of the cylinder, and R is the radius of the cylinder.

5. The method for self-evaluating the accuracy of the UAV laser point cloud based on the data itself according to claim 1, characterized in that In step (5), the calculation formula of the mean value m of the distances from each projection point to point p is as follows: In the above formula, N is the number of points in the cylindrical neighborhood C(P) of point P, and d i is the distance from the projection of the i-th point in C(P) onto the central axis of the cylinder to the projection p of point P, and m is the average value of the projection distances; The method for evaluating the accuracy of point P according to the mean value m of the distances from each projection point to point p is:

Images