A photovoltaic support spindle recognition method, system and device based on radar point cloud
By combining radar point cloud PCA principal component analysis and RANSAC fitting with multi-frame accumulation and task-driven screening, the problems of adhered point clouds and sparse point clouds in the identification of photovoltaic support main shafts were solved, realizing accurate identification and real-time pose calculation of photovoltaic support main shafts, adapting to various terrain changes.
Patent Information
- Application Number
- CN202611260015.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-08-19
- Publication Date
- 2026-11-03
AI Technical Summary
Existing technologies for identifying the main shaft of photovoltaic brackets have the problem of difficulty in effectively separating the sticky point cloud from the main shaft and other structures such as purlins and columns. Furthermore, the point cloud in a single frame is sparse and has high uncertainty, resulting in insufficient recognition accuracy and poor robustness. In particular, it is difficult to achieve real-time pose calculation in complex environments and undulating terrain.
Interference objects are filtered out by using principal component analysis (PCA) of radar point clouds. By combining RANSAC multiple fitting approximation and threshold filtering methods with multi-frame accumulation and task-driven point cloud screening, non-principal axis points are gradually eliminated, the principal axis direction is determined, and geometric parameters are calculated to output reliable pose information.
It enables accurate identification and real-time pose calculation of the main shaft of photovoltaic support in complex environments, reduces system deployment and maintenance costs, improves the reliability and accuracy of identification, and adapts to various terrain changes.
Smart Images
Figure CN123033017A_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of photovoltaic installation technology, and more specifically, to a method, system, and device for identifying the main shaft of a photovoltaic support based on radar point clouds. Background Technology
[0002] In the automated installation of photovoltaic (PV) modules, a module installation robot is needed to install the modules one by one onto the PV support structure. The installation process involves two steps: first, a robotic arm moves the module above the main shaft for coarse positioning; second, sensors at the end of the robotic arm detect the purlins for secondary positioning, completing the precise installation. The PV support structure consists of a main shaft, purlins, columns, and fasteners connected together. In relatively flat PV sites, the height of the main shaft from the ground does not change significantly, and the coarse positioning height of the installation robot can be considered constant. However, in hilly or undulating PV sites, the height of the main shaft varies greatly, and a fixed value is not suitable. Since the PV main shaft is a horizontal column tens of meters above the ground, using LiDAR point cloud detection to identify the main shaft is a feasible method. However, the main shaft is not a single optical axis; it is connected to purlins, columns, fasteners, etc., and there may be other objects nearby, such as tall vegetation, which poses a challenge to LiDAR point cloud identification. Summary of the Invention
[0003] To address the aforementioned technical problems, this application discloses a method, system, and device for identifying the main shaft of a photovoltaic support system based on radar point clouds. It captures the horizontal and elongated characteristics of the main shaft, filters the data using point cloud PCA principal component analysis, and then approximates it through multiple RANSAC fittings and threshold filtering to ensure accurate and reliable main shaft identification by lidar. Specifically, the technical solution of this application is as follows: In a first aspect, this application discloses a method for identifying the main shaft of a photovoltaic support based on radar point clouds, comprising the following steps: Point cloud data of the target area of the photovoltaic support is collected using radar sensors; Based on the installation task type of the component installation robot and the prior position range of the main axis relative to the component installation robot, the regions of interest of the target area point cloud are clustered and filtered in the XYZ three dimensions to obtain candidate main axis point clouds; The candidate principal axis point cloud is fitted with a straight line fitting algorithm multiple times. The distance threshold used in each fitting gradually decreases as the number of fittings increases, so as to eliminate non-principal axis points in the candidate principal axis point cloud and finally determine the principal axis direction straight line and the corresponding valid point cloud. The geometric parameters of the candidate principal axes are calculated based on the principal axis direction line and the effective points, and the geometric parameters are verified based on preset validity judgment conditions to obtain valid principal axis data. Based on the main axis direction line and the effective main axis data, the pose information of the target point on the main axis relative to the component installation robot is calculated.
[0004] In some embodiments, the acquisition of point cloud data of the target area of the photovoltaic support using radar sensors specifically includes: The photovoltaic support structure is equipped with a non-repetitive scanning radar to collect multi-frame point cloud data of the area, and the multi-frame point cloud data is accumulated and superimposed to obtain the point cloud of the target area.
[0005] In some implementations, the step of clustering and filtering the regions of interest (ROIs) of the target area point cloud in the XYZ dimensions based on the installation task type of the component installation robot and the prior position range of the main axis relative to the component installation robot to obtain candidate main axis point clouds specifically includes: Based on the current task of the component installation robot, determine whether the component to be installed is located on the left or right side of the component installation robot, and retain the point cloud of the target area on the corresponding side and remove the point cloud of the non-operation area on the opposite side. Based on the prior position range of the main axis relative to the component installation robot, spatial thresholds are set in the three dimensions of the robot coordinate system: the X-axis, Y-axis, and Z-axis. The target region point cloud is cropped to the region of interest including the main axis to obtain the region of interest point cloud. The point cloud of the region of interest is clustered and segmented to obtain several point cloud clusters. Then, each point cloud cluster is filtered based on preset filtering conditions to obtain candidate principal axis point clouds.
[0006] In some implementations, the step of clustering and segmenting the point cloud of the region of interest to obtain several point cloud clusters, and then filtering each point cloud cluster based on preset filtering conditions to obtain candidate principal axis point clouds, specifically includes: Voxel filtering is used to downsample the point cloud of the region of interest to obtain sparse point cloud data. The sparse point cloud data is segmented using Euclidean clustering to obtain several independent point cloud clusters; Principal component analysis is performed on each of the point cloud clusters to calculate the eigenvalues and eigenvectors of the covariance matrix of each point cloud cluster. The direction of the eigenvector corresponding to the maximum eigenvalue is determined as the principal direction of the corresponding point cloud cluster. Based on the preset filtering conditions, each point cloud cluster is filtered to obtain the point cloud cluster that satisfies the filtering conditions in the main direction as the candidate main axis point cloud.
[0007] In other embodiments, the preset filtering conditions include: The angle between the main direction of the point cloud clustering and the preset coordinate axis direction is less than a first threshold. The length of the point cloud cluster along the main direction is greater than the second threshold. The angle between the main direction and the vertical direction of the point cloud clustering is greater than the third threshold.
[0008] In some implementations, the geometric parameters of the candidate principal axis are calculated based on the principal axis direction line and the effective point; specifically including: Based on the main axis direction line and the effective point cloud, calculate the projection value of each point in the effective point cloud on the main axis direction line, and obtain the length of the main axis by subtracting the maximum projection value from the minimum projection value; Calculate the spatial vertical distance from each point in the effective point cloud to the straight line in the direction of the main axis, and multiply the maximum value of the spatial vertical distance by two to obtain the width of the main axis; The pitch angle of the main axis is obtained by calculating the angle between the direction vector of the main axis direction line and the horizontal plane.
[0009] In some implementations, the geometric parameters are verified based on preset validity criteria to obtain valid spindle data; specifically including: The candidate principal axes whose geometric parameters satisfy the validity judgment conditions are taken as valid principal axes; the geometric parameters corresponding to the valid principal axes are taken as the valid principal axis data; If the geometric parameters do not meet the validity judgment conditions, then the identification result of the candidate principal axis is discarded. Optionally, the validity judgment conditions include: The length of the main shaft is greater than the fourth threshold. The ratio of the length to the width of the spindle is greater than a fifth threshold. The absolute value of the pitch angle of the main shaft is less than the sixth threshold.
[0010] In other embodiments, based on the main axis direction line and the effective main axis data, the pose information of the target point on the main axis relative to the component installation robot is calculated; specifically including: Define a specific plane perpendicular to the main axis X direction in the robot coordinate system; the specific plane is X=A; where A is the x-coordinate value of the target point along the main axis X direction in the robot coordinate system; Calculate the spatial intersection of the specific plane and the straight line of the main axis direction to obtain the first pose of the target point in the sensor coordinate system; By transforming the coordinates of the radar sensor relative to the robot body on which the component is installed, the first pose of the target point in the sensor coordinate system is transformed to the robot coordinate system, thus obtaining the second pose information of the target point in the robot coordinate system; the second pose information includes height, distance, pitch angle, and rotation angle data.
[0011] Secondly, this application also discloses a photovoltaic support main shaft identification system based on radar point cloud, including a memory, a processor, and a computer program stored in the memory. The processor executes the computer program to implement the steps of the photovoltaic support main shaft identification method based on radar point cloud described in any of the above embodiments.
[0012] Thirdly, this application also discloses a component installation robot device, the device including the photovoltaic bracket spindle recognition system based on radar point cloud described in the above embodiments.
[0013] Compared with the prior art, this application has at least one of the following beneficial effects: 1. This application identifies the principal axis based on the geometric features of radar point clouds. It obtains the principal direction through principal component analysis of the point cloud, identifies elongated structures through screening conditions, and achieves precise positioning through multiple straight-line fittings. The entire algorithm relies solely on the geometric properties of the principal axis, not its external appearance. Therefore, this invention eliminates the need for model training sample collection and can adaptively identify elongated structures with square, circular, semi-circular, or even other cross-sections. It eliminates the need to change the algorithm or retrain the model, significantly reducing system deployment and maintenance costs.
[0014] 2. This application constructs a complete technical chain. It effectively filters out non-main axis interference through Euclidean clustering and point cloud PCA principal component analysis, and then accurately separates adhered point clouds using a progressive RANSAC fitting strategy with gradually tightening distance thresholds. The final output is pose information that can be directly used for robot control. This achieves fully automatic, real-time main axis recognition and pose calculation, possessing high practical value and promising prospects for widespread application.
[0015] 3. This application uses LiDAR as the sensing sensor, which can stably collect high-precision point cloud data even in complex lighting environments such as strong outdoor light, backlight, and alternating shadows. Furthermore, based on the characteristics of a relatively slender main axis, where a single-frame radar scan results in sparse point clouds on the main axis with high uncertainty, this application solves the problem of sparse point clouds on a slender main axis through multi-frame accumulation and reduces computational complexity through voxel filtering. This effectively ensures the installation robot's continuous and reliable identification of the main axis. Attached Figure Description
[0016] The preferred embodiments will now be described in a clear and easy-to-understand manner, in conjunction with the accompanying drawings, to further explain the above-mentioned characteristics, technical features, advantages, and implementation methods of this application.
[0017] Figure 1 This is a flowchart illustrating the steps of an embodiment of a photovoltaic support main shaft identification method based on radar point clouds according to this application. Figure 2 This is a typical schematic diagram of the photovoltaic support main shaft in the embodiments of this application; Figure 3 This is a schematic diagram of the component installation robot installation operation in an embodiment of this application; Figure 4 This is a flowchart illustrating a sub-step of step S2 in an embodiment of this application. Figure 5 This is a flowchart illustrating another embodiment of a photovoltaic support main shaft identification method based on radar point clouds according to this application; Figure 6 This is a schematic block diagram of an embodiment of a photovoltaic support spindle identification system based on radar point cloud according to this application. Detailed Implementation
[0018] In the following description, specific details such as particular system architectures and techniques are set forth for illustrative purposes and not for limitation, in order to provide a thorough understanding of the embodiments of this application. However, those skilled in the art will understand that this application can also be implemented in other embodiments without these specific details. In other instances, detailed descriptions of well-known systems, apparatuses, circuits, and methods have been omitted so as not to obscure the description of this application with unnecessary detail.
[0019] It should be understood that, when used in this specification and the appended claims, the term "comprising" indicates the presence of the described features, integrals, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components and / or sets.
[0020] To keep the drawings concise, each figure only schematically shows the parts relevant to the invention, and these do not represent the actual structure of the product. Furthermore, to facilitate understanding, in some figures, only one of components with the same structure or function is schematically depicted, or only one is labeled. In this document, "one" not only means "only one," but can also mean "more than one."
[0021] It should also be further understood that the term “and / or” as used in this application specification and the appended claims means any combination of one or more of the associated listed items and all possible combinations, and includes such combinations.
[0022] In this document, it should be noted that, unless otherwise explicitly specified and limited, the terms "installation," "connection," and "linking" should be interpreted broadly. For example, they can refer to fixed connections, detachable connections, or integral connections; they can refer to mechanical connections or electrical connections; they can refer to direct connections or indirect connections through an intermediate medium; and they can refer to the internal connection between two components. Those skilled in the art can understand the specific meaning of the above terms in this application based on the specific circumstances.
[0023] Furthermore, in the description of this application, the terms "first," "second," etc., are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.
[0024] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the specific implementation methods of this application will be described below with reference to the accompanying drawings. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings and other implementation methods can be obtained based on these drawings without creative effort.
[0025] In recent years, with the continuous growth of global demand for clean energy, the scale of photovoltaic power plant construction has been expanding. In the field of photovoltaic module installation, the application of intelligent construction equipment is becoming increasingly widespread. Automated installation of photovoltaic modules through artificial intelligence and computer vision technology can reduce module installation time by up to 50%. Currently, existing automated installation technologies for photovoltaic installation robots typically consist of two stages: coarse positioning and fine positioning. The robot moves the module to a rough position above the main shaft, and then uses end-effector sensors to detect the purlins for precise installation. In relatively flat areas, the main shaft's height above the ground is relatively stable, and coarse positioning can use fixed height parameters. However, in hillsides or complex terrain with significant undulations, the main shaft height varies significantly, making traditional fixed-value methods difficult to adapt. Therefore, sensors such as LiDAR are needed to perceive the main shaft's position and attitude in real time.
[0026] Despite some progress in existing technologies, significant limitations remain in the specific application scenario of photovoltaic support shaft recognition. First, most existing LiDAR sensing methods target relatively "clean" target objects, while the photovoltaic support shaft is not an isolated structure. It is tightly connected to purlins, columns, fasteners, and other components, and may also contain obstructions such as tall vegetation. These structures are intertwined with the shaft in the point cloud, making them difficult to separate effectively using traditional clustering methods. Second, as a slender cylindrical structure, the number of points falling on the shaft in a single LiDAR scan is extremely limited. The point cloud is sparse and highly uncertain, making direct processing of single-frame data prone to recognition failure or insufficient accuracy. Existing methods rarely offer effective solutions for the sparse point cloud problem of slender targets. Third, existing LiDAR sensing technology has poor robustness in extreme weather (such as heavy fog, rain, snow, and strong light) or complex environments, and the sensed data is easily interfered with. This limits the robot's ability to perceive the features of the work object. In addition, existing methods lack technical solutions for situations where the main axis height changes continuously under undulating terrain, and can adaptively identify the main axis position and output real-time pose information that can be used for coarse robot localization.
[0027] This application addresses the aforementioned technical deficiencies by developing an effective spindle identification system. It captures the horizontal and elongated characteristics of the spindle, filters the data using point cloud PCA principal component analysis, and then applies multiple RANSAC fitting approximations and threshold filtering to ensure accurate and reliable spindle identification by the LiDAR. (See attached specification.) Figure 1 As shown, an embodiment of a photovoltaic support main shaft identification method based on radar point clouds according to this application specifically includes the following steps: S1, collect point cloud data of the target area of the photovoltaic support using a radar sensor.
[0028] For details, please refer to the attached instruction manual. Figure 2 As shown, Figure 2 This is a typical schematic diagram of a photovoltaic support main shaft, which is relatively slender with a diameter of approximately 10-25 cm (the main shaft that can be identified in this application is not limited to this diameter range). A single-frame radar scan results in sparse point clouds falling on the main shaft, and the presence or absence of point clouds at the shaft's edges is highly uncertain, causing interference with identification. Accumulating multiple frames can effectively improve the point cloud density and continuity in the main shaft region, reducing the adverse effects of random noise and missing data in a single frame on subsequent identification.
[0029] Therefore, in some implementations, multi-frame point cloud data of the area where the photovoltaic support is located is collected by a non-repetitive scanning radar, for example, 5-10 frames of point cloud data are collected, and the multi-frame point cloud data are accumulated and superimposed to obtain the point cloud of the target area.
[0030] In other alternative implementations, using a lidar with a higher line count can also overcome the problem of a sparse number of main axis point clouds and unstable edge distribution acquired by a single frame scan.
[0031] S2, based on the installation task type of the component installation robot and the prior position range of the main axis relative to the component installation robot, cluster and filter the regions of interest of the target area point cloud in the XYZ three dimensions to obtain candidate main axis point clouds.
[0032] Specifically, in the initial screening step, firstly, based on the current task of the photovoltaic module installation robot, it is determined whether the module to be installed is located on the left or right side of the robot. Point cloud data in the corresponding lateral direction is retained, while point cloud data from non-operational areas on the opposite side is discarded. Then, based on the prior position range of the main axis relative to the installation robot, spatial thresholds are set in the X, Y, and Z axes of the robot coordinate system. The point cloud data is then cropped to a target cuboid region containing the main axis, resulting in the region of interest point cloud. This step, through task-driven lateral screening and three-dimensional spatial limitation, effectively reduces the data scale for subsequent processing while eliminating a large number of environmental interference point clouds unrelated to the main axis.
[0033] S3, a straight line fitting algorithm is used to perform multiple straight line fittings on the candidate principal axis point cloud. The distance threshold used in each fitting gradually decreases as the number of fittings increases, so as to eliminate non-principal axis points in the candidate principal axis point cloud, and finally determine the principal axis direction straight line and the corresponding valid point cloud.
[0034] For details, please refer to Figure 2 Since the purlins, columns, and fasteners are all tightly connected to the shaft, the point cloud at this time often contains a point cloud that integrates the main shaft point cloud scanned by the radar with the point clouds of the purlins, columns, and fasteners. However, this application only needs to obtain the center line of the main shaft point cloud to determine the pose of the main shaft.
[0035] This application employs the Random Sample Consensus (RANSAC) algorithm to perform multiple iterations of straight line fitting on the candidate principal axis point cloud. The principle of RANSAC straight line fitting is to draw a straight line connecting two points in the point cloud, calculate the number of points in the point cloud whose distance from the straight line is within a distance threshold, and iterate to find the straight line with the largest number of points as the result.
[0036] Because of interference from purlins, pillars, and other structures in the point cloud, the straight lines obtained by RANSAC will tilt towards the purlins and pillars. Therefore, this application employs a straight line fitting algorithm to perform multiple straight line fittings on the candidate principal axis point cloud, wherein the distance threshold used in each fitting gradually decreases as the number of fittings increases. In specific implementation, a larger distance threshold is set during the first fitting to include outliers in the candidate principal axis point cloud that deviate from the main body of the principal axis due to adhesion from auxiliary structures such as purlins, pillars, and fasteners, thus initially obtaining the approximate direction of the principal axis.
[0037] In each subsequent iteration, the distance threshold is gradually decreased by a preset step size, causing the fitted line to gradually approach the core point cloud region of the main axis, and successively eliminating auxiliary structure point clouds that deviate significantly from the main axis. After several iterations, the fitted line converges to the central axis of the main axis, and a set of effective point clouds whose distance from this line is within the final threshold range is obtained.
[0038] This embodiment provides some other optional implementation methods, which employ a multi-cycle RANSAC linear fitting algorithm. The initial value of the RANSAC distance threshold is relatively large, such as twice the size of the spindle diameter. Then, the RANSAC distance threshold is set to gradually decrease with the number of cycles, such as using 10 cycles, each time decreasing by 0.1 times the spindle diameter, gradually converging to obtain the spindle center.
[0039] S4. The geometric parameters of the candidate principal axis are calculated based on the principal axis direction line and the effective points, and the geometric parameters are verified based on the preset validity judgment conditions to obtain the effective principal axis data.
[0040] Specifically, after obtaining the principal axis direction lines and valid point clouds through multiple RANSAC line fittings, the geometric parameters of candidate principal axes are first calculated based on the valid point clouds. The geometric parameters of the principal axes include the principal axis length, principal axis width, and principal axis direction vector. A second validity assessment is then performed based on these geometric parameters to discard invalid assessment results.
[0041] S5. Based on the main axis direction line and the effective main axis data, calculate the pose information of the target point on the main axis relative to the component installation robot.
[0042] For details, please refer to the attached instruction manual. Figure 3 As shown, Figure 3 This is a schematic diagram of the component installation robot's installation operation in this embodiment. After obtaining valid spindle data, the system further calculates the pose information of the target point (e.g., position A) on the spindle relative to the component installation robot. This includes the target point's height, horizontal distance, pitch angle, and rotation angle in the robot coordinate system, forming complete pose information output, which serves as the control basis for the component installation robot to perform coarse positioning operations on the components.
[0043] Based on the above embodiments, this application discloses another embodiment of a photovoltaic support main shaft identification method based on radar point clouds, wherein step S2 specifically includes the following sub-steps: S21, based on the current task of the component installation robot, determine whether the component to be installed is located on the left or right side of the component installation robot, and retain the point cloud of the target area in the corresponding side direction, and remove the point cloud of the non-operation area on the opposite side.
[0044] Specifically, during the automated installation of photovoltaic modules, the module installation robot typically installs modules one by one in a left-to-right or right-to-left order. Therefore, the control system clearly knows whether the module to be installed is located on the left or right side of the module installation robot. Based on this task information, the system performs lateral filtering on the multi-frame point cloud data accumulated in step S1.
[0045] Specifically, when the task is left-side installation, the system uses the origin of the robot coordinate system as a reference, retaining the point cloud data of the left region and discarding the point cloud data of the right region. When the task is right-side installation, the system retains the point cloud data of the right region and discards the point cloud data of the left region. This lateral filtering operation can effectively reduce the amount of data processed subsequently, while eliminating interfering point clouds from other photovoltaic support structures, construction equipment, or personnel in the non-operational area on the other side, thus improving the targeting and accuracy of spindle recognition.
[0046] S22, based on the prior position range of the main axis relative to the component installation robot, spatial thresholds are set in the X, Y, and Z dimensions of the robot coordinate system. The target region point cloud is cropped to the region of interest including the main axis to obtain the region of interest point cloud.
[0047] Specifically, after completing the lateral screening, the system further performs three-dimensional spatial trimming based on the prior position range of the main axis relative to the installation robot. Since the main axis of the photovoltaic bracket has certain regularities in the installation site, such as the height of the main axis from the ground usually being within a preset range, the horizontal distance between the main axis and the robot being limited by the travel range of the installation operation, and the offset of the main axis in the Y-axis direction being limited by the lateral position of the component installation, reasonable spatial thresholds can be set in the three dimensions of the robot coordinate system: X-axis, Y-axis, and Z-axis.
[0048] Specifically, the system determines the minimum and maximum values in the X-axis, Y-axis, and Z-axis directions based on the robot's current position, preset installation station coordinates, and typical spindle dimensions, forming a cuboid-shaped region of interest. Then, all points falling outside this cuboid region in the point cloud data obtained in step S21 are removed, retaining only the point cloud within the region to obtain the region of interest point cloud. This operation significantly reduces the spatial range of point cloud processing and effectively eliminates irrelevant interference from distant vegetation, other photovoltaic support structures, and other obstructions.
[0049] S23, perform clustering and segmentation on the point cloud of the region of interest to obtain several point cloud clusters, and then filter each point cloud cluster based on preset filtering conditions to obtain candidate principal axis point clouds.
[0050] Optionally, the system performs clustering and principal component analysis (PCA) on the point cloud of the region of interest. This separates spatially isolated point cloud clusters into several independent clusters, and calculates the principal direction and extension length along the principal direction for each cluster. Finally, based on preset screening criteria, each cluster is judged, and interfering clusters that deviate from the expected direction, have excessively short axial lengths, or are too vertical are eliminated. The cluster that best matches the long axis characteristics is selected as the candidate principal axis point cloud.
[0051] This application provides another embodiment of a photovoltaic support main shaft identification method based on radar point clouds. Based on any embodiment of the above method, refer to the appendix to the specification. Figure 4 As shown, step S23 involves clustering and segmenting the point cloud of the region of interest to obtain several point cloud clusters. Then, based on preset filtering conditions, each point cloud cluster is filtered to obtain candidate principal axis point clouds. Specifically, this includes: S231, the voxel filtering method is used to perform voxel filtering downsampling on the point cloud of the region of interest to obtain sparse point cloud data.
[0052] Specifically, after obtaining the point cloud of the region of interest, the system first performs voxel filtering on the point cloud, dividing the three-dimensional space into a voxel grid of a specified size, retaining only one representative point in each voxel, thereby achieving downsampling and uniform distribution of the point cloud and reducing the computational complexity of subsequent calculations.
[0053] S232, Euclidean clustering is performed on the sparse point cloud data to obtain several independent point cloud clusters.
[0054] Specifically, the Euclidean clustering algorithm is used to segment the voxel-filtered point cloud. Based on the Euclidean distance between points, spatially separated point cloud clusters are distinguished, resulting in several independent point cloud clusters. Each cluster corresponds to an independent object or part of an object in space.
[0055] S233, perform principal component analysis on each of the point cloud clusters, calculate the eigenvalues and eigenvectors of the covariance matrix of each point cloud cluster, and determine the direction of the eigenvector corresponding to the maximum eigenvalue as the principal direction of the corresponding point cloud cluster.
[0056] Specifically, the system performs principal component analysis on each point cloud cluster, calculates the eigenvalues and eigenvectors of the covariance matrix, and determines the direction of the eigenvector corresponding to the maximum eigenvalue as the principal direction of the cluster.
[0057] S234, based on the preset filtering conditions, each point cloud cluster is filtered to obtain the point cloud cluster that satisfies the filtering conditions in the main direction as the candidate main axis point cloud.
[0058] Specifically, based on the typical spatial attitude characteristics of photovoltaic spindles in installation scenarios, the system sets multiple screening conditions.
[0059] Optionally, in some implementations, the preset filtering conditions include: The angle between the main direction of the point cloud clustering and the preset coordinate axis direction is less than a first threshold.
[0060] The length of the point cloud cluster along the main direction is greater than the second threshold.
[0061] The angle between the main direction and the vertical direction of the point cloud clustering is greater than the third threshold.
[0062] In practice, all point cloud clusters are traversed, and clusters that do not meet any of the conditions are eliminated. Among the clusters that meet all conditions, the cluster with the longest projection length in the principal direction is selected as the candidate principal axis point cloud. Through the above steps, the system can effectively filter out the candidate point clouds that are most likely to be the principal axis from point clouds in complex environments.
[0063] Based on the above embodiments, this application discloses another embodiment of a photovoltaic support main axis identification method based on radar point clouds. In step S3, after performing multiple straight line approximation fittings on the optimal point cloud to obtain the main axis direction straight line and the effective point cloud, step S4 is continued. The geometric parameters of the candidate main axis are calculated based on the main axis direction straight line and the effective point cloud, and the geometric parameters are verified based on preset validity judgment conditions to obtain the effective main axis data. For the complete process, please refer to the attached specification. Figure 5 As shown.
[0064] In other embodiments of this example, step S4 specifically includes the following sub-steps: S41, based on the main axis direction line and the effective point cloud, calculate the projection value of each point in the effective point cloud on the main axis direction line, and obtain the length of the main axis by subtracting the maximum projection value from the minimum projection value.
[0065] Specifically, the calculation method for the principal axis length is as follows: Project all points in the valid point cloud onto the straight line of the principal axis direction, and obtain a projection value for each point. Traverse all valid point clouds to find the maximum and minimum projection values, and calculate the principal axis length as the maximum projection value minus the minimum projection value.
[0066] S42, calculate the spatial vertical distance from each point in the effective point cloud to the straight line of the main axis direction, and multiply the maximum value of the spatial vertical distance by two to obtain the width of the main axis.
[0067] Specifically, the calculation method for the main axis width is as follows: calculate the spatial vertical distance from each point in the effective point cloud to the straight line in the direction of the main axis, traverse all effective point clouds to find the maximum value among these distance values, and this maximum value reflects the maximum distance from the center of the main axis to the edge.
[0068] S43, calculate the angle between the direction vector of the main axis direction line and the horizontal plane to obtain the pitch angle of the main axis.
[0069] Specifically, the unit direction vector is extracted from the fitted principal axis direction line, denoted as (Vx, Vy, Vz), where Vz represents the vertical component of the principal axis. Since the normal vector of the horizontal plane is (0,0,1), the angle between the principal axis direction and the horizontal plane can be obtained by calculating the vertical component of the direction vector, specifically using the formula: Pitch angle = arctan(Vz / sqrt(Vx² + Vy²)). This yields the angle between the principal axis direction vector and its projection onto the horizontal plane.
[0070] Optionally, when the spindle is completely horizontal, Vz=0 and the pitch angle is 0; when the spindle pitches up or down, Vz is not zero and the pitch angle is positive or negative respectively.
[0071] In some implementations, step S4 further includes: S44, the candidate principal axes whose geometric parameters satisfy the validity judgment condition are taken as valid principal axes. The geometric parameters corresponding to the valid principal axes are taken as the valid principal axis data.
[0072] S45, if the geometric parameters do not meet the validity judgment conditions, then the identification result of the candidate principal axis is discarded.
[0073] The validity judgment criteria include: The length of the main shaft is greater than the fourth threshold.
[0074] The ratio of the length to the width of the spindle is greater than a fifth threshold.
[0075] The absolute value of the pitch angle of the main shaft is less than the sixth threshold.
[0076] Specifically, the system compares the calculated spindle length with a preset length threshold, the ratio of spindle length to spindle width with a preset aspect ratio threshold, and the pitch angle of the spindle direction with a preset pitch angle threshold. Only when all three conditions are met—spindle length greater than the length threshold, length-to-width ratio greater than the aspect ratio threshold, and absolute pitch angle less than the pitch angle threshold—is the current recognition result considered valid, and valid spindle data obtained. If any condition is not met, the recognition is considered a failure, and the current result is discarded.
[0077] In other implementations, the spindle length must meet a fourth threshold; for example, the spindle length needs to be greater than or equal to 1.5 meters. The ratio of the spindle length to the spindle width must be greater than a fifth threshold, for example, set to greater than or equal to 5. The absolute value of the spindle pitch angle must be less than a sixth threshold; for example, the absolute value of the spindle pitch angle can be set to be less than 20 degrees. If all three conditions are met, the fitted spindle direction line is considered correct.
[0078] This application provides another embodiment of a photovoltaic support main shaft identification method based on radar point clouds. Based on any embodiment of the above method, step S5 specifically includes the following sub-steps: S51, determine a specific plane perpendicular to the main axis X direction in the robot coordinate system. The specific plane is X=A, where A is the x-coordinate value of the target point along the main axis X direction in the robot coordinate system.
[0079] Specifically, firstly, based on the process requirements of the installation task, a specific plane perpendicular to the X-axis of the main axis is determined in the robot coordinate system, X=A; the constant A represents the coordinate value of the target installation position in front of the robot on the X-axis, refer to the appendix of the instruction manual. Figure 3 As shown.
[0080] S52, calculate the spatial intersection of the specific plane and the main axis direction line, that is, obtain the first pose of the target point in the sensor coordinate system.
[0081] Specifically, it is necessary to calculate the pose of the main axis A. Based on the sensor's installation pose on the robot, it is only necessary to extract the intersection point of the plane X=A and the main axis direction line, the direction of which is determined by the main axis direction line.
[0082] S53, by transforming the coordinates of the radar sensor relative to the robot body on which the component is installed, the first pose of the target point in the sensor coordinate system is transformed to the robot coordinate system, thus obtaining the second pose information of the target point in the robot coordinate system. The second pose information includes height, distance, pitch angle, and rotation angle data.
[0083] Specifically, through a pre-defined coordinate transformation relationship between the radar sensor and the component installation robot, the fitted main axis direction line is transformed from the sensor coordinate system to the robot coordinate system. In the robot coordinate system, the spatial intersection point of a specific plane X equal to a constant A and the transformed main axis direction line is calculated. This intersection point represents the position coordinates of the target point on the main axis in the robot coordinate system. Simultaneously, the pitch and rotation angles of the main axis are determined based on the direction vector of the main axis direction line. The position coordinates and attitude information together constitute the complete six-degree-of-freedom pose of the target point in the robot coordinate system, including height, distance, pitch angle, and rotation angle data. Finally, this pose information is output to the motion control system of the installation robot to guide the robotic arm in performing a coarse positioning operation of the component upwards along the main axis.
[0084] Based on the same concept, this application also discloses a photovoltaic support main shaft identification system based on radar point clouds. It includes a memory, a processor, and a computer program stored in the memory. The processor executes the computer program to implement the steps of the photovoltaic support main shaft identification method based on radar point clouds described in any of the above embodiments.
[0085] In some implementations, the system includes at least one radar sensor mounted on the actuator of the component installation robot for acquiring three-dimensional point cloud data of the working environment.
[0086] It also includes a processor module. See attached reference manual. Figure 6As shown. The processor module includes at least the following types of execution units: For example, a preprocessing unit, used to accumulate multiple frames of point cloud data and perform voxel filtering downsampling on the accumulated point cloud to solve the sparsity problem of the main axis point cloud and reduce the data volume; a clustering and segmentation unit, used to perform Euclidean clustering on the preprocessed point cloud to obtain several independent point cloud clusters; a principal component analysis unit, used to calculate the principal direction and the extension length along the principal direction for each point cloud cluster; a candidate selection unit, used to select each cluster based on preset geometric conditions to obtain candidate main axis point clouds; a line fitting unit, used to perform line fitting on the candidate main axis point clouds using a RANSAC algorithm with gradually decreasing distance thresholds to obtain the main axis direction line and valid point clouds; a parameter verification unit, used to calculate the geometric parameters of the main axis based on the main axis direction line and valid point clouds and perform validity judgment to obtain valid main axis data; and a pose calculation unit, used to calculate the pose information of the target point in the robot coordinate system based on the valid main axis data and a preset installation position plane.
[0087] Based on the same concept, this application also discloses a component installation robot device, the device including the photovoltaic bracket spindle recognition system based on radar point cloud described in the above embodiments.
[0088] The photovoltaic support main shaft identification method, system and device based on radar point cloud in this application have the same technical concept, and the technical details of the embodiments of the three are applicable to each other. In order to reduce repetition, they will not be repeated here.
[0089] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the above-described division of program modules is merely an example. In practical applications, the above functions can be assigned to different program modules as needed, that is, the internal structure of the device can be divided into different program units or modules to complete all or part of the functions described above. The program modules in the embodiments can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one processing unit. The integrated unit can be implemented in hardware or as a software program unit. Furthermore, the specific names of the program modules are only for easy differentiation and are not intended to limit the scope of protection of this application.
[0090] Obviously, those skilled in the art can make various modifications and variations to this application without departing from the spirit and scope of this application. Therefore, if such modifications and variations fall within the scope of the claims of this application and their equivalents, this application also intends to include such modifications and variations.
Claims
1. A method for identifying the main shaft of a photovoltaic support based on radar point clouds, characterized in that, Includes the following steps: Point cloud data of the target area of the photovoltaic support is collected using radar sensors; Based on the installation task type of the component installation robot and the prior position range of the main axis relative to the component installation robot, the regions of interest of the target area point cloud are clustered and filtered in the XYZ three dimensions to obtain candidate main axis point clouds; The candidate principal axis point cloud is fitted with a straight line fitting algorithm multiple times. The distance threshold used in each fitting gradually decreases as the number of fittings increases, so as to eliminate non-principal axis points in the candidate principal axis point cloud and finally determine the principal axis direction straight line and the corresponding valid point cloud. The geometric parameters of the candidate principal axes are calculated based on the principal axis direction line and the effective points, and the geometric parameters are verified based on preset validity judgment conditions to obtain valid principal axis data. Based on the main axis direction line and the effective main axis data, the pose information of the target point on the main axis relative to the component installation robot is calculated.
2. The photovoltaic support main shaft identification method based on radar point cloud as described in claim 1, characterized in that, The acquisition of point cloud data of the target area of the photovoltaic support using radar sensors specifically includes: The photovoltaic support structure is equipped with a non-repetitive scanning radar to collect multiple frames of point cloud data in the area where the photovoltaic support structure is located, and the point cloud of the target area is obtained by accumulating and superimposing the multiple frames of point cloud data.
3. The photovoltaic support main shaft identification method based on radar point cloud as described in claim 1, characterized in that, The method of clustering and filtering the regions of interest (ROIs) of the target area point cloud in the X, Y, and Z dimensions based on the installation task type of the component installation robot and the prior position range of the main axis relative to the component installation robot to obtain candidate main axis point clouds specifically includes: Based on the current task of the component installation robot, determine whether the component to be installed is located on the left or right side of the component installation robot, and retain the point cloud of the target area on the corresponding side and remove the point cloud of the non-operation area on the opposite side. Based on the prior position range of the main axis relative to the component installation robot, spatial thresholds are set in the three dimensions of the robot coordinate system: the X-axis, Y-axis, and Z-axis. The target region point cloud is cropped to the region of interest including the main axis to obtain the region of interest point cloud. The point cloud of the region of interest is clustered and segmented to obtain several point cloud clusters. Then, each point cloud cluster is filtered based on preset filtering conditions to obtain candidate principal axis point clouds.
4. The photovoltaic support main shaft identification method based on radar point cloud as described in claim 3, characterized in that, The process of clustering and segmenting the point cloud of the region of interest to obtain several point cloud clusters, and then filtering each point cloud cluster based on preset filtering conditions to obtain candidate principal axis point clouds, specifically includes: Voxel filtering is used to downsample the point cloud of the region of interest to obtain sparse point cloud data. The sparse point cloud data is segmented using Euclidean clustering to obtain several independent point cloud clusters; Principal component analysis is performed on each of the point cloud clusters to calculate the eigenvalues and eigenvectors of the covariance matrix of each point cloud cluster. The direction of the eigenvector corresponding to the maximum eigenvalue is determined as the principal direction of the corresponding point cloud cluster. Based on the preset filtering conditions, each point cloud cluster is filtered to obtain the point cloud cluster that satisfies the filtering conditions in the main direction as the candidate main axis point cloud.
5. The photovoltaic support main shaft identification method based on radar point cloud as described in claim 4, characterized in that, The preset filtering conditions include: The angle between the main direction of the point cloud clustering and the preset coordinate axis direction is less than a first threshold. The length of the point cloud cluster along the main direction is greater than the second threshold. The angle between the main direction and the vertical direction of the point cloud clustering is greater than the third threshold.
6. The photovoltaic support main shaft identification method based on radar point cloud as described in claim 1, characterized in that, The geometric parameters of the candidate principal axis are calculated based on the principal axis direction line and the effective points; specifically including: Based on the main axis direction line and the effective point cloud, calculate the projection value of each point in the effective point cloud on the main axis direction line, and obtain the length of the main axis by subtracting the maximum projection value from the minimum projection value; Calculate the spatial vertical distance from each point in the effective point cloud to the straight line in the direction of the main axis, and multiply the maximum value of the spatial vertical distance by two to obtain the width of the main axis; The pitch angle of the main axis is obtained by calculating the angle between the direction vector of the main axis direction line and the horizontal plane.
7. The photovoltaic support main shaft identification method based on radar point cloud as described in claim 6, characterized in that, The geometric parameters are validated based on preset validity criteria to obtain valid spindle data; specifically including: The candidate principal axes whose geometric parameters satisfy the validity judgment conditions are taken as valid principal axes; the geometric parameters corresponding to the valid principal axes are taken as the valid principal axis data; If the geometric parameters do not meet the validity judgment conditions, then the identification result of the candidate principal axis is discarded. The validity judgment criteria include: The length of the main shaft is greater than the fourth threshold. The ratio of the length to the width of the spindle is greater than a fifth threshold. The absolute value of the pitch angle of the main shaft is less than the sixth threshold.
8. The photovoltaic support main shaft identification method based on radar point cloud as described in claim 1, characterized in that, Based on the main axis direction line and the effective main axis data, calculate the pose information of the target point on the main axis relative to the component installation robot; specifically including: Define a specific plane perpendicular to the main axis X direction in the robot coordinate system; the specific plane is X=A; where A is the x-coordinate value of the target point along the main axis X direction in the robot coordinate system; Calculate the spatial intersection of the specific plane and the straight line of the main axis direction to obtain the first pose of the target point in the sensor coordinate system; By transforming the coordinates of the radar sensor relative to the robot body on which the component is installed, the first pose of the target point in the sensor coordinate system is transformed to the robot coordinate system, thus obtaining the second pose information of the target point in the robot coordinate system; the second pose information includes height, distance, pitch angle, and rotation angle data.
9. A photovoltaic support spindle identification system based on radar point clouds, comprising a memory, a processor, and a computer program stored in the memory, characterized in that, The processor executes the computer program to implement the steps of the photovoltaic support spindle identification method based on radar point cloud as described in any one of claims 1-8.
10. A component assembly robot device, characterized in that, The device includes the photovoltaic support spindle identification system based on radar point cloud as described in claim 9.