A point cloud acquisition system based on lidar and a pan-tilt head and an insulator recognition and positioning method

Abstract

This application relates to a point cloud acquisition system and an insulator identification and positioning method based on a lidar and a pan-tilt head. The point cloud acquisition system includes a lidar, a computer, and a pan-tilt head. The lidar is vertically installed at the exact center of the pan-tilt head. The computer is respectively connected to the lidar and the pan-tilt head. The computer drives the lidar to scan through ROS, and at the same time drives the pan-tilt head to rotate uniformly at a fixed angle, so as to realize the scanning of the entire scene. After scanning, the lidar transmits the scanned point cloud data to the computer through the Ethernet protocol for subsequent processing. This application directly uses three-dimensional point clouds for insulator identification, and can directly output the spatial coordinate information of the insulators while completing the identification task, with a faster processing speed and stronger pertinence.

Description

Technical Field

[0001] This application relates to the technical field of insulator identification and positioning, and particularly to a point cloud acquisition system and an insulator identification and positioning method based on a lidar and a pan-tilt head. Background Art

[0002] Insulators play a role in supporting conductors and insulating in fields such as substations and transmission lines. Therefore, it is required to have good mechanical properties and electrical properties. Since insulators are subject to long-term effects of temperature changes, mechanical stress, air pollution, etc. during operation, it is necessary to regularly inspect and clean insulators to ensure their reliable operation. However, the method of manual live working has high requirements for the technical level and proficiency of operators, weather conditions, safety protection equipment, etc. Robot live inspection and cleaning has become the trend of the development of power grid technology. The first problem to be solved is the identification and positioning of insulators in the working space.

[0003] Existing insulator identification methods use images as a medium to identify insulator information in images through image processing algorithms. However, it is difficult to directly give the spatial position information of insulators, and other methods need to be used to solve the spatial position information of insulators to guide the movement of robots, with high computational complexity. In the identification and positioning of insulators in other fields, deep learning methods are used, but deep learning methods require a large amount of sample data for training, consuming a large amount of cost, and it is also difficult to directly output the spatial position information of insulators. Summary of the Invention

[0004] The purpose of the embodiments of this application is to provide a point cloud acquisition system and an insulator identification and positioning method based on a lidar and a pan-tilt head, which directly use three-dimensional point clouds to identify insulators, and can directly output the spatial coordinate information of insulators while completing the identification task, with faster processing speed and stronger pertinence.

[0005] To achieve the above purpose, this application provides the following technical solutions:

[0006] In a first aspect, the embodiments of this application provide a point cloud acquisition system based on a lidar and a pan-tilt head, including a lidar, a computer, and a pan-tilt head.

[0007] The lidar is vertically installed at the center of the pan-tilt head.

[0008] The computer is respectively connected to the lidar and the pan-tilt head.

[0009] The computer drives the lidar to scan through ROS, and at the same time drives the pan-tilt head to rotate at a constant speed by a fixed angle, so as to realize the scanning of the entire scene. After scanning, the lidar transmits the scanned point cloud data to the computer through the Ethernet protocol for subsequent processing.

[0010] The lidar is a 360-degree lidar, and both the lidar and the pan-tilt are disc-shaped structures.

[0011] In a second aspect, an insulator recognition and positioning method based on a lidar and a pan-tilt provided by an embodiment of the present application includes the following specific steps:

[0012] Start the lidar and the pan-tilt to scan a large-scale scene;

[0013] Filter the point cloud obtained by scanning based on prior knowledge to obtain the region of interest;

[0014] Use the region growing method to perform clustering analysis on the point cloud in the region of interest to obtain different clustering clusters;

[0015] Calculate the viewpoint histogram features of different clustering clusters;

[0016] Calculate the Euclidean distance between the viewpoint histogram features of different clustering clusters and the insulator template;

[0017] Complete the recognition and positioning of the insulator through a given threshold and display it in the scanned scene.

[0018] The specific method for filtering the point cloud obtained by scanning to obtain the region of interest is to first process the original point cloud through the method of direct filtering. The method of direct filtering establishes a channel based on the point cloud coordinate system, filters out the point cloud outside the channel, and only retains the point cloud inside the channel. The point cloud inside the channel represents the region of interest.

[0019] The specific method for using the region growing method to perform clustering analysis on the point cloud in the region of interest to obtain different clustering clusters is to adopt a segmentation algorithm based on region growing. By calculating the curvature of all points in the point cloud and sorting them, select the point with the smallest curvature as the initial seed point. Then design an empty seed point sequence and an empty clustering array, and set the number of neighboring points for seed point search. Starting from the initial seed point, if the angle threshold between the normal of the neighboring point and the normal of the current seed point is less than the given threshold, add the neighboring point to the current clustering array. Then check the curvature of the current neighboring point. If the curvature is less than the given threshold, add the neighboring point to the seed point sequence and delete the current seed point, and continue to grow with the new seed point until the seed point sequence is empty, and one region growth is completed. Finally, repeat the above operations for the remaining points until all points are traversed.

[0020] The insulator is identified and located by a given threshold and displayed in the scanned scene. The identification of the insulator specifically involves traversing the clustering clusters, calculating the viewpoint histogram features of each clustering point cloud, and calculating the Euclidean distance between the viewpoint histogram features of the clustering point cloud and those of the insulator template. If the Euclidean distance is less than the given threshold, the clustering point cloud is considered an insulator.

[0021] The positioning of the insulator and its display in the scanned scene specifically involve enveloping the clustering point cloud identified as an insulator using a bounding box, calculating the coordinates of its center point in the point cloud coordinate system, and displaying it in the original point cloud, thereby achieving the positioning and display of the insulator.

[0022] Compared with the prior art, the beneficial effects of the present invention are as follows: Compared with the traditional insulator identification method based on image processing, the media used for identification and positioning in this method are different. By using a lidar and a pan-tilt unit for three-dimensional scanning of a large scene, the insulator is directly identified using the scanned three-dimensional lidar point cloud, and the spatial position information of the insulator can be directly output; compared with the identification method based on deep learning, this method does not require a large amount of manpower and material resources to obtain and classify the data set, and can achieve the identification and positioning of the insulator at a relatively long distance with high identification accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the following will briefly introduce the drawings required to be used in the embodiments of the present application. It should be understood that the following drawings only show some embodiments of the present application, and therefore should not be regarded as limiting the scope. For those of ordinary skill in the art, other related drawings can be obtained based on these drawings without creative efforts.

[0024] Figure 1 It is a point cloud acquisition system diagram of an embodiment of the present application;

[0025] Figure 2 It is a flow chart of the insulator identification and positioning method of the present application;

[0026] Figure 3 It is an original point cloud diagram of large-scale scene reconstruction in an embodiment of the present application;

[0027] Figure 4 It is a point cloud diagram after direct filtering of the x-axis of the original point cloud of the present application;

[0028] Figure 5 It is a point cloud diagram after direct filtering of the y-axis of the original point cloud of the present application;

[0029] Figure 6 It is a flow chart of the region-growing based segmentation algorithm of the present application;

[0030] Figure 7 is the flow chart of the insulator recognition algorithm of this application;

[0031] Figure 8 is the front view of the insulator positioning effect in a large-scale scene of this application;

[0032] Figure 9 is the perspective view of the insulator positioning effect in a large-scale scene of this application;

[0033] Figure 10 is the center point coordinate of the envelope box of this application. Specific implementation manner

[0034] Next, the technical solutions in the embodiments of this application will be described in conjunction with the accompanying drawings in the embodiments of this application. It should be noted that: similar reference numerals and letters represent similar items in the following drawings. Therefore, once an item is defined in one drawing, it does not need to be further defined and explained in subsequent drawings.

[0035] The term "including", "comprising" or any other variant thereof is intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article or device. Without further limitation, an element defined by the statement "including a..." does not exclude the presence of additional identical elements in the process, method, article or device including the element.

[0036] In the first aspect, as Figure 1 shown, the present invention provides a point cloud acquisition system based on a lidar and a pan-tilt head, including a lidar 1, a computer 2, and a pan-tilt head 3,

[0037] The lidar 1 is vertically installed at the exact center of the pan-tilt head 3;

[0038] The computer 2 is respectively connected to the lidar 1 and the pan-tilt head 3;

[0039] The computer 2 drives the lidar 1 to scan through ROS. The computer 2 simultaneously drives the pan-tilt head 3 to rotate at a constant speed at a fixed angle, so as to achieve the scanning of the entire scene; after scanning, the lidar 1 transmits the scanned point cloud data to the computer 2 through the Ethernet protocol for subsequent processing.

[0040] The lidar 1 is a 360-degree lidar, and both the lidar 1 and the pan-tilt head 3 are of a disc-shaped structure.

[0041] As Figure 2As shown in the second aspect, an insulator recognition and positioning method based on a lidar and a pan-tilt head is provided in an embodiment of the present application, including the following specific steps:

[0042] Start the lidar and the pan-tilt head to scan a large-scale scene;

[0043] Filter the scanned point cloud based on prior knowledge to obtain the region of interest;

[0044] Use the region growing method to perform clustering analysis on the point cloud in the region of interest to obtain different clustering clusters;

[0045] Calculate the view point histogram features of different clustering clusters;

[0046] Calculate the Euclidean distance between the view point histogram features of different clustering clusters and the insulator template;

[0047] Complete the recognition and positioning of the insulator by a given threshold and display it in the scanned scene.

[0048] As Figure 3 、 Figure 4 and Figure 5 shown, Figure 3 For the original point cloud of the large-scale scene reconstruction, since the lidar scans the entire three-dimensional scene, the number of original point clouds is 962,639, and the point cloud information is very redundant and contains a large number of uninteresting regions. Since the position distance between the lidar and the insulator is relatively fixed, a method of direct filtering is used to process the original point cloud first. The method of direct filtering establishes a channel based on the point cloud coordinate system, filters out the point cloud outside the channel and only retains the point cloud inside the channel. The point cloud inside the channel represents the region of interest, and the number of its point clouds is greatly reduced compared to the number of original point clouds. The present invention filters the original point cloud on the x-axis and y-axis. The channel range of the x-axis is (3m, 7m). The point cloud diagram after the x-axis direct filtering is as Figure 4 shown, the number of point clouds is 54,766, and the insulators have been manually marked. The y-axis range is (-2m, 2m). The point cloud diagram after the y-axis direct filtering is as Figure 5 shown, the number of point clouds is 15,390, and the insulators have been manually marked.

[0049] After performing direct filtering on the original point cloud, the point cloud data is greatly reduced. However, the point cloud data at this time is still disorderly. Therefore, it is necessary to perform clustering analysis on the filtered point cloud to classify the point cloud of the insulator into the same category. The present invention adopts a region-growing segmentation algorithm. By calculating the curvature of all points in the point cloud and sorting them, the point with the smallest curvature is selected as the initial seed point. Then, an empty seed point sequence and an empty clustering array are designed, and the number of neighborhood points for seed point search is set. Starting from the initial seed point for search, if the angle threshold between the normal of the neighborhood point and the normal of the current seed point is less than the given threshold, the neighborhood point is added to the current clustering array. Then, the curvature of the current neighborhood point is checked. If the curvature is less than the given threshold, the neighborhood point is added to the seed point sequence and the current seed point is deleted, and growth continues with the new seed point until the seed point sequence is empty, indicating that one region growth is completed. Finally, the above operations are repeated for the remaining points until all points are traversed. The flowchart of the region-growing segmentation algorithm is as Figure 6 shown, where the number of neighborhood points for seed point search is 30, the normal angle threshold is π / 60, and the curvature threshold is 1.

[0050] From the above steps, 21 point cloud clustering clusters can be obtained. By traversing the clustering clusters, the viewpoint histogram features of each clustering point cloud are calculated, and the Euclidean distance is calculated between them and the viewpoint histogram features of the insulator template. The viewpoint histogram features consist of the fast point feature histogram and the viewpoint component. Assume that each clustering point cloud is represented by the point set P, the s-th point of the point cloud is p s , the normal vector of point p s is n s , the t-th neighboring point of point p s is p t , the normal vector of point p t is n t , and a coordinate system uvw is defined on point p s , and its basis vectors are:

[0051] u = n s

[0052]

[0053] w = u × v

[0054] Based on the coordinate system uvw, the normal vector difference n s between point p t and point p j can be described by a set of feature point features SPFH = (α, β, γ, d), and its calculation formula is:

[0055] α = cos -1 (v · n j )

[0056]

[0057] γ = tan -1 (w·n j , u·n j )

[0058] d = ‖p s - p t ‖²

[0059] For a point p in the point set s , it has k neighboring points. By weighting its own point features with those of its k neighboring points, its fast point features can be obtained. The calculation formula is as follows:

[0060]

[0061] In the formula, w i represents the distance between point p s and its i-th neighboring point, and k = 50.

[0062] For each component in the FPFH of each point in the point set P, it is divided into 45 intervals. By counting the number of times each component appears in the corresponding 45 intervals and then plotting its histogram, a 180-dimensional fast point feature histogram vector can be obtained.

[0063] The view point component is obtained by calculating the angle θ between the view point v of the point set P p and each point p s on it, and dividing it into 128 intervals. By counting the number of times each component appears in the 128 intervals and then plotting its histogram, a 128-dimensional view point component histogram can be obtained.

[0064] By combining the 180-dimensional fast point histogram and the 128-dimensional view point component histogram, a 308-dimensional view point histogram feature VFH can be obtained.

[0065] Obtain the view point histogram feature VFH of the clustered point cloud i and the view point histogram feature VFH of the template point cloud template , and calculate the Euclidean distance d between the two feature vectors:

[0066]

[0067] If the Euclidean distance is less than the given threshold, then the clustered point cloud is considered to be an insulator. The algorithm flow chart for insulator recognition is as Figure 7 shown, where the Euclidean distance threshold is 100.

[0068] The clustered point clouds identified as insulators above are enveloped using a bounding box, and the coordinates of the center point in the point cloud coordinate system are calculated and displayed in the original point cloud, so as to realize the positioning and display of the insulators. The results are as Figure 8 and Figure 9 shown. Among them, the bounding box represents the position of the located insulator, and the three mutually perpendicular lines represent the origin and construction method of the current point cloud coordinate system. The coordinates of the center point of the bounding box are as Figure 10 shown.

[0069] In the embodiments of this application, a point cloud acquisition system based on lidar and a pan-tilt is designed, which can effectively cover the hemispherical environmental space, can realize large-scale three-dimensional scanning of substation and distribution network environments, and finally realize the point cloud acquisition of multiple insulators, effectively solving the problem of the small reconstruction range of traditional lidar;

[0070] In the embodiments of this application, a method for directly identifying and positioning insulators using three-dimensional point clouds is proposed, which can accurately identify insulators and output their spatial position information;

[0071] In the embodiments of this application, the prior information of distance is proposed to complete the rough positioning of insulators. Compared with the overall positioning of other methods, the amount of point cloud processing is less and the efficiency is higher;

[0072] In the embodiments of this application, a three-dimensional point cloud template of insulators is proposed to complete the rapid matching and positioning of insulators in the point cloud. Compared with the method of using a dataset to train a model, this method is more targeted and more feasible in practical applications;

[0073] In the embodiments of this application, a bounding box is proposed to complete the spatial positioning of insulators. Compared with other methods that are easily affected by sparse point cloud data when detecting the edges of insulators, this method can completely extract the spatial position and attitude of insulators, and the success rate is higher.

[0074] The key content of this invention is that, different from the traditional image-based insulator recognition method, this invention proposes a new point cloud acquisition and insulator recognition and positioning method, which directly uses three-dimensional point clouds to recognize insulators, and can directly output the spatial coordinate information of insulators while completing the recognition task, with faster processing speed and stronger pertinence.

[0075] The above are only the embodiments of this application and are not used to limit the protection scope of this application. For those skilled in the art, this application can have various changes and modifications. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of this application shall be included in the protection scope of this application.

Claims

1. An insulator recognition and positioning method based on lidar and a pan-tilt head, characterized in that The specific steps include: Start the laser radar and gimbal to scan a large range of scenes; Based on prior knowledge, the scanned point cloud is filtered to obtain the region of interest; The region growing method is used to perform cluster analysis on the point cloud of the region of interest to obtain different clusters; Calculate the viewpoint histogram features of different clusters; Calculate the Euclidean distance between the viewpoint histogram features of different clusters and the insulator template; The insulator is identified and located by a given threshold and displayed in the scanned scene; The specific method of filtering the scanned point cloud to obtain the region of interest is to first process the original point cloud by a straight-through filtering method, wherein the straight-through filtering method establishes a channel based on the point cloud coordinate system, filters out the point cloud outside the channel and only retains the point cloud within the channel, and the point cloud within the channel represents the region of interest; The method of using the region growing method to perform cluster analysis on the point cloud of the region of interest to obtain different cluster clusters is as follows: a segmentation algorithm based on region growing is used, the curvature of all points in the point cloud is calculated and sorted, the point with the smallest curvature is selected as the initial seed point, and then an empty seed point sequence and an empty cluster array are designed, and the number of domain points for seed point search is set, and the search starts with the initial seed point. If the angle threshold between the normal of the domain point and the normal of the current seed point is less than a given threshold, the domain point is added to the current cluster array, and then the curvature of the current domain point is checked. If the curvature is less than a given threshold, the domain point is added to the seed point sequence and the current seed point is deleted, and the growth continues with the new seed point until the seed point sequence is empty and a region growth is completed. Finally, the above operation is repeated for the remaining points until all points are traversed.

2. The insulator recognition and positioning method based on lidar and pan-tilt according to claim 1, characterized in that, The identification and positioning of the insulator is completed by a given threshold and displayed in the scanned scene, wherein the identification of the insulator is specifically performed by traversing the cluster clusters, calculating the viewpoint histogram features of each cluster point cloud, and performing Euclidean distance calculation on the viewpoint histogram features of the insulator template. If the Euclidean distance is less than the given threshold, the cluster point cloud is considered to be an insulator.

3. The insulator recognition and positioning method based on lidar and pan-tilt according to claim 2, characterized in that, The positioning of the insulator and displaying it in the scanned scene specifically includes enveloping the clustered point cloud identified as the insulator with a bounding box, calculating the coordinates of its center point in the point cloud coordinate system, and displaying it in the original point cloud, thereby realizing the positioning and display of the insulator.

4. A point cloud acquisition system based on a lidar and a pan-tilt head, for implementing the method according to claim 1, characterized in that, It includes a laser radar (1), a computer (2) and a gimbal (3). The laser radar (1) is vertically mounted at the exact center of the pan / tilt platform (3); The computer (2) is connected to the laser radar (1) and the pan-tilt platform (3) respectively; The computer (2) drives the laser radar (1) to scan via ROS, and the computer (2) simultaneously drives the pan / tilt platform (3) to rotate at a fixed angle at a constant speed, thereby realizing scanning of the entire scene; after scanning, the laser radar (1) transmits the scanned point cloud data to the computer (2) via an Ethernet protocol for subsequent processing.

5. The point cloud acquisition system based on lidar and pan-tilt according to claim 4, characterized in that, The laser radar (1) is a 360-degree laser radar, and the laser radar (1) and the pan-tilt platform (3) are both disc-shaped structures.

Images