Helicopter rotor blade with aerofoil section

Abstract

The application belongs to the technical field of heliostat pose estimation and anomaly detection method, and particularly relates to a heliostat pose estimation and anomaly detection method based on unmanned aerial vehicle reconstructed point cloud, ground points are extracted from the unmanned aerial vehicle reconstructed point cloud, a ground surface reference is established based on the ground points, HAG height normalization is performed on non-ground points, and mirror surface candidate point clusters are obtained through height threshold screening; PCA plane fitting is performed on the mirror surface candidate point clusters, an initial center, a normal vector and low-quality candidates are obtained, a pentagon heliostat geometric prior is introduced, a parameterized pentagon template is constructed, a center position and an in-plane orientation angle are used as optimization variables, a mirror surface pose {c, n, theta} is obtained through joint optimization based on a target function, wherein c is the center position, n is the normal vector, and theta is the in-plane orientation angle.

Description

Technical Field

[0001] This invention belongs to the technical field of heliostat pose estimation and anomaly detection methods, and particularly relates to a heliostat pose estimation and anomaly detection method based on point cloud reconstruction by UAV. Background Technology

[0002] Traditional heliostat pose calibration and anomaly detection mainly rely on manual target measurement and contact-based inspection, which suffers from low inspection efficiency, high maintenance costs, and an inability to achieve full coverage inspection of large-scale heliostat fields. Furthermore, manual operation is susceptible to environmental and human interference, and the accuracy and timeliness of pose data acquisition cannot meet the needs of automated, high-frequency maintenance in solar thermal power plants. Existing point cloud-based heliostat pose calculation methods are mostly adapted to LiDAR point clouds, exhibiting poor adaptability to point clouds reconstructed by UAV photogrammetry. They are easily affected by factors such as uneven point cloud density, the presence of holes and defects, and undulating terrain within the heliostat field. Additionally, they lack specific geometric prior constraints for pentagonal heliostats, resulting in insufficient pose estimation accuracy and robustness. Anomaly detection is also prone to false anomalies at boundaries and low recognition accuracy, failing to provide reliable support for efficient heliostat field maintenance and precise correction. Summary of the Invention

[0003] In view of the aforementioned problems, and in conjunction with the first aspect of the present invention, embodiments of the present invention provide a method for heliostat pose estimation and anomaly detection based on point cloud reconstruction from an unmanned aerial vehicle (UAV), the method comprising: Includes the following steps: Step 1: Extract ground points from the point cloud reconstructed by the UAV, establish a surface reference based on the ground points, perform HAG height normalization on non-ground points, and obtain mirror candidate point clusters based on height thresholds; Step 2: Perform PCA plane fitting on the candidate point clusters of the mirror surface to obtain the initial center and normal vector, and filter out low-quality candidates; introduce the geometric prior of the pentagonal heliostat, construct a parameterized pentagonal template, and use the center position and in-plane orientation angle as optimization variables to jointly optimize the mirror pose {c,n,theta} based on the objective function, where c is the center position, n is the normal vector, and theta is the in-plane orientation angle; Step 3: Calculate the weighted average normal angle between each mirror and its k nearest neighbors, mark mirrors with angles exceeding the anomaly threshold as anomalous mirrors, and perform weight reduction processing on the mirrors at the edge of the sub-region to reduce false anomalies.

[0004] Preferably, in step one, a continuous surface reference is established based on ground point interpolation or Delaunay triangulation, and HAG height normalization is based on the local ground as the zero reference, thereby reducing the systematic influence of the topographic undulation of the mirror field on height selection.

[0005] Preferably, in step two, the side length parameter of the pentagonal heliostat geometric prior is 5.3 meters, and the circumscribed circle radius parameter is 4.508 meters. The tolerance zone is set with the nominal size to accommodate installation errors, point cloud missing and local deviations caused by boundary clipping.

[0006] Preferably, in step two, the objective function is: F(c,θ)=Nin(c,θ)+β·Nedge(c,θ)-γ·Nout(c,θ); Where Nin is the number of points inside the template, Nedge is the number of points on the template boundary, Nout is the number of points penalized outside the template, and β and γ are weight coefficients; The optimization stopping condition is that the change in the center position is less than a threshold τ for two consecutive iterations. c And the change in orientation angle is less than the threshold τ θ .

[0007] Preferably, in step two, the candidate point cluster is first projected onto the PCA fitting plane to establish a local two-dimensional coordinate system, with the center and principal direction of the PCA fitting as the initial optimization values.

[0008] Preferably, in step three, k is set to 8, at which point the anomaly detection accuracy and recall are optimal.

[0009] Preferably, in step three, the abnormal threshold is set to 70 mrad; Instead of directly removing edge mirrors, we use a weighting reduction approach to preserve their statistical participation ability and reduce their impact on the overall discrimination results.

[0010] Preferably, in step two, the mirror pose {c,n,theta} is used to evaluate the spot offset pattern, quantify energy loss, and guide the heliostat mechanical servo system to accurately correct its deviation.

[0011] Preferably, in step two, the fitting residual and the number of points threshold are used to filter low-quality mirror candidate point clusters.

[0012] Based on the above, relying on UAV aerial surveying enables non-contact, high-frequency, and rapid inspection of the heliostat field, completely eliminating the drawbacks of traditional manual target calibration, such as low efficiency, high cost, and inability to cover a large area. Employing HAG height normalization and pentagonal geometric prior constraints, it effectively overcomes the problems of uneven point cloud density and hole defects in UAV-reconstructed points, significantly improving the accuracy and robustness of heliostat pose estimation in complex occlusion scenarios. The mirror pose parameters, jointly optimized by the center and orientation angles, can intuitively reflect the heliostat's focusing state, providing reliable data support for spot offset assessment, energy loss quantification, and servo system correction. The anomaly detection method based on neighborhood consistency exhibits excellent accuracy and recall, accurately identifying discrete abnormal mirror surfaces. It achieves high success rate and low false anomaly rate in large-scale mirror field extraction, greatly simplifying on-site operation and maintenance review and decision-making processes. Attached Figure Description

[0013] Figure 1 This invention relates to an abnormal mirror spatial distribution; Figure 2 The present invention relates to an anomalous spatial distribution of tilt angles; Figure 3 This is the neighborhood normal difference distribution of the present invention; Figure 4 This is the result of the tilt angle clustering in this invention; Figure 5 This is the spatial clustering result of the present invention; Figure 6 This is an example of a three-dimensional view of the abnormal area in this invention. Detailed Implementation

[0014] The present invention will now be described in detail with reference to the accompanying drawings. Figure 1 This is a flowchart illustrating a method for heliostat pose estimation and anomaly detection based on point cloud reconstruction from an unmanned aerial vehicle (UAV) according to an embodiment of the present invention. The following is a detailed description of this method for heliostat pose estimation and anomaly detection based on point cloud reconstruction from an UAV.

[0015] This invention provides a method for reconstructing point clouds from unmanned aerial vehicles (UAVs), which consists of three parts: data preprocessing and candidate extraction, pose estimation and pentagonal constraint, and neighborhood consistency anomaly detection. The method emphasizes "reconstructed point cloud adaptability" and "pentagonal mirror features" to avoid over-reliance on laser point cloud conditions.

[0016] Data preprocessing and candidate extraction begin with HAG (Height Above Ground) height normalization using a ground model. Preferably, ground points are extracted from the point cloud first, and a surface reference is established based on ground point interpolation or a Delaunay triangulation. Then, the relative ground height is calculated for non-ground points. Compared to directly using the original elevation values ​​for height thresholding, HAG uses the local ground as a zero reference, which can weaken the systematic influence of topographic undulations on the selection of mirror height zones. Compared to the discrete normalization method of pixel-by-pixel subtraction using a raster elevation model, the continuous normalization method based on point-by-point interpolation of ground points or a Delaunay triangulation can reduce the normalization error caused by topographic discretization. Public data shows that after simple subtraction of a raster DTM, the normalized ground points may still have a residual deviation of approximately ±25cm, while point-by-point normalization of the point cloud is superior to discrete topographic subtraction in terms of computational accuracy. Therefore, this invention uses HAG height normalization as a preliminary step for candidate mirror height selection.

[0017] Pose estimation and pentagonal constraints are used to perform PCA plane fitting on candidate clusters to obtain the center and normal vectors. Low-quality candidates are filtered out using the fitting residuals and a point count threshold. Based on this, a pentagonal heliostat geometric prior is introduced: the mirror coverage area is constrained using a side length parameter of 5.3 meters, a circumcircle radius parameter of 4.508 meters, and the distribution of points along the edge band. Joint optimization of the center position and orientation angle is performed to maximize the number of points inside the pentagon and correct the initial fitting offset. The mirror pose is represented by a triple {c, n, theta}, where c is the center position, n is the normal vector, and theta is the in-plane orientation angle. In the actual search, an allowable tolerance zone is set with the nominal size as the center, so that the geometric prior can reflect the true size of the mirror while tolerating local deviations caused by installation errors, missing point clouds, and boundary clipping. The candidate cluster points are projected onto the fitting plane to establish a local two-dimensional coordinate system; using the center and principal direction obtained from PCA plane fitting as initial values, a parameterized pentagonal template M(c, θ) is constructed. Let Nin(c,θ) be the number of points inside the template, Nedge(c,θ) be the number of points inside the template boundary band, and Nout(c,θ) be the number of penalty points outside the template. Then, by solving the objective function... F(c,θ)=Nin(c,θ)+β·Nedge(c,θ)-γ·Nout(c,θ); The center position c and orientation angle θ are jointly optimized, where β and γ are weighting coefficients. The optimization stopping condition can be set as follows: the change in center position is less than a threshold τ in two consecutive iterations. c And the change in orientation angle is less than the threshold τ θ This optimization method does not simply minimize the distance from a point to a plane, but directly utilizes the pentagonal mirror boundary coverage relationship to improve the consistency between the candidate mirror and the actual mirror profile.

[0018] Neighborhood consistency anomaly detection: In a regular array, the local mirror normals should have continuity; therefore, anomalous mirrors can be identified through neighborhood consistency. This method calculates the angle difference between a mirror and the average normal of its k nearest neighbors, marking mirrors with differences exceeding a threshold as anomalous. To stabilize the statistical caliber, mirrors at the edges of sub-regions are removed or downweighted to reduce false anomalies caused by clipping boundaries. For the normal vector n of the i-th mirror... i Search for the k nearest neighbor mirrors in space and calculate the weighted neighborhood average normal vector m. i and the angle s between the two i =arccos(n i ·m i)As an abnormal indicator. To reduce the statistical bias caused by the clipping and occlusion of the sub-region boundaries, the neighborhood contribution of the edge mirrors is down-weighted instead of being directly excluded, enabling the edge mirrors to still retain the ability to participate in the statistics, while reducing their impact on the global discrimination result. Through comparative experiments, the accuracy / recall rate of abnormal detection is optimal when k = 8. When k < X, the neighborhood statistics are insufficient, and when k > X, it is vulnerable to the pseudo-abnormal interference of occluded mirrors.

[0019] The present invention proposes a method for pentagon heliostat pose estimation and abnormal detection based on unmanned aerial vehicle (UAV) reconstructed point clouds, achieving the following remarkable effects in practical engineering applications. Efficient inspection and low-cost coverage: It completely breaks through the limitations of traditional manual target calibration, which is inefficient, costly, and difficult to cover a large area. It realizes non-contact and high-frequency rapid inspection of large-scale heliostat fields by using UAV aerial survey. Precise calculation and defect overcoming: It effectively overcomes the interference of non-uniform point cloud density and hole defects in UAV photogrammetry. By innovatively introducing pentagon geometric prior constraints for joint optimization, it greatly improves the pose estimation accuracy in complex occlusion environments. Data support and operation and maintenance closed-loop (the role of pose): The accurately calculated pose parameters {c, n, theta} directly reflect the real-time focusing state of the heliostat, providing reliable data support for subsequent evaluation of the spot offset law, quantification of energy loss, and guidance of precise rectification of the mechanical servo system.

[0020] Based on the visual reconstruction three-dimensional model collected at the UAV flight altitude of 100 m, the upper limit of the accuracy of mirror normal estimation based on the visual reconstruction model is verified in the local experimental area. The results of single-mirror multi-level downsampling show that when the number of sampled points is 5, 20, 100, and all 499 points respectively, the angular standard deviations obtained by RANSAC fitting are 6.44 mrad, 1.50 mrad, 0.54 mrad, and 0.20 mrad respectively; the average residual of the points to the fitting plane under full data is about 0.50 cm. Among the 12 mirrors selected in the horizontal area, the average deviation of the mirror group is 7.78 mrad, the maximum deviation is 12.27 mrad, and the standard deviation is 3.07 mrad; in the inclined area, the average deviation of the mirror group is 9.60 mrad, and the maximum deviation is 13.18 mrad; in the abnormal area, 1 significantly abnormal mirror is successfully identified, with a deviation value of 347.0 mrad. After excluding the abnormal mirror, the average deviation of the normal mirrors is 9.92 mrad. The above results show that the mirror pose estimation based on the UAV visual reconstruction model is feasible and can provide an accuracy verification basis for subsequent automated point cloud processing algorithms.

[0021] A large-scale experiment was conducted within approximately 1 / 8 of the point cloud region. Based on a total of 1687 mirrors, after removing 11 edge mirrors, 1676 core heliostats were successfully extracted. Of these, 1599 mirrors had complete parameter extraction, achieving a success rate of 95.41%. Within an anomaly threshold of 70 mrad, 3 anomalous mirrors were detected (e.g., ...). Figure 1 As shown in the figure), the anomaly rate is 0.18%. Experiments show that the pose calculations for most mirror surfaces are stable, and the number of detected anomalous mirror surfaces is small, with a discrete spatial distribution. Related visualization results (such as...) Figures 1 to 6 As shown, it can clearly and intuitively reveal the spatial absolute position and relative tilt angle differences of abnormal mirrors. The high concentration of tilt angle distribution and stable neighborhood consistency index make the real abnormal mirrors exhibit highly identifiable outlier characteristics, which greatly facilitates the rapid verification and maintenance decisions of on-site engineers.

[0022] It should be noted that, in order to simplify the description of the present invention and thus help to understand one or more embodiments of the invention, multiple features may sometimes be grouped into one embodiment, drawing or description thereof in the foregoing description of the embodiments of the present invention.

[0023] The embodiments of this application have been described above with reference to the accompanying drawings. Unless otherwise specified, the embodiments and features in the embodiments of this application can be combined with each other. This application is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms under the guidance of this application without departing from the spirit and scope of the claims, and all of these forms are within the protection scope of this application.

Claims

1. A method for heliostat pose estimation and anomaly detection based on point cloud reconstruction by UAV, characterized in that: Includes the following steps: Step 1: Extract ground points from the point cloud reconstructed by the UAV, establish a surface reference based on the ground points, perform HAG height normalization on non-ground points, and obtain mirror candidate point clusters based on height thresholds; Step 2: Perform PCA plane fitting on the candidate point clusters of the mirror surface to obtain the initial center and normal vector, and filter out low-quality candidates; introduce the geometric prior of the pentagonal heliostat, construct a parameterized pentagonal template, and use the center position and in-plane orientation angle as optimization variables to jointly optimize the mirror pose {c,n,theta} based on the objective function, where c is the center position, n is the normal vector, and theta is the in-plane orientation angle; Step 3: Calculate the weighted average normal angle between each mirror and its k nearest neighbors, mark mirrors with angles exceeding the anomaly threshold as anomalous mirrors, and perform weight reduction processing on the mirrors at the edge of the sub-region to reduce false anomalies.

2. The method for heliostat pose estimation and anomaly detection based on UAV-reconstructed point cloud as described in claim 1, characterized in that: In step one, a continuous surface reference is established based on ground point interpolation or Delaunay triangulation. HAG height normalization uses the local ground as the zero reference, which weakens the systematic influence of the topographic relief of the mirror field on height selection.

3. The method for heliostat pose estimation and anomaly detection based on UAV-reconstructed point cloud as described in claim 1, characterized in that: In step two, the side length parameter of the pentagonal heliostat geometric prior is 5.3 meters, and the circumcircle radius parameter is 4.508 meters. The tolerance zone is set with the nominal size to accommodate installation errors, point cloud missing and local deviations caused by boundary clipping.

4. The method for heliostat pose estimation and anomaly detection based on UAV-reconstructed point cloud as described in claim 1, characterized in that: In step two, the objective function is: F(c,θ)=Nin(c,θ)+β·Nedge(c,θ)-γ·Nout(c,θ); Where Nin is the number of points inside the template, Nedge is the number of points on the template boundary, Nout is the number of points penalized outside the template, and β and γ are weight coefficients; The optimization stopping condition is that the change in the center position is less than a threshold τ for two consecutive iterations. c And the change in orientation angle is less than the threshold τ θ .

5. The method for heliostat pose estimation and anomaly detection based on UAV-reconstructed point cloud as described in claim 1, characterized in that: In step two, the candidate point cluster is first projected onto the PCA fitting plane to establish a local two-dimensional coordinate system, with the center and principal direction of the PCA fitting as the initial values ​​for optimization.

6. The method for heliostat pose estimation and anomaly detection based on UAV-reconstructed point cloud as described in claim 1, characterized in that: In step three, k is set to 8, at which point the anomaly detection accuracy and recall are optimal.

7. The method for heliostat pose estimation and anomaly detection based on UAV-reconstructed point cloud as described in claim 1, characterized in that: In step three, the anomaly threshold is set to 70 mrad; Instead of directly removing edge mirrors, we use a weighting reduction approach to preserve their statistical participation ability and reduce their impact on the overall discrimination results.

8. The method for heliostat pose estimation and anomaly detection based on UAV-reconstructed point cloud as described in claim 1, characterized in that: In step two, the mirror pose {c,n,theta} is used to evaluate the spot offset pattern, quantify energy loss, and guide the heliostat mechanical servo system to accurately correct its deviation.

9. The method for heliostat pose estimation and anomaly detection based on UAV-reconstructed point cloud as described in claim 1, characterized in that: In step two, the fitting residual and the number of points threshold are used to filter low-quality mirror candidate point clusters.

Images