CN112785711B - Insulator creepage distance detection method and detection system based on three-dimensional reconstruction - Google Patents
Insulator creepage distance detection method and detection system based on three-dimensional reconstruction Download PDFInfo
- Publication number
- CN112785711B CN112785711B CN202110083707.3A CN202110083707A CN112785711B CN 112785711 B CN112785711 B CN 112785711B CN 202110083707 A CN202110083707 A CN 202110083707A CN 112785711 B CN112785711 B CN 112785711B
- Authority
- CN
- China
- Prior art keywords
- point cloud
- insulator
- cloud data
- sampling
- creepage distance
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
- 239000012212 insulator Substances 0.000 title claims abstract description 110
- 238000001514 detection method Methods 0.000 title claims abstract description 41
- 238000005070 sampling Methods 0.000 claims abstract description 84
- 238000004422 calculation algorithm Methods 0.000 claims abstract description 47
- 238000000034 method Methods 0.000 claims abstract description 29
- 238000009499 grossing Methods 0.000 claims abstract description 18
- 230000008569 process Effects 0.000 claims abstract description 11
- 238000012545 processing Methods 0.000 claims abstract description 10
- 239000013598 vector Substances 0.000 claims description 21
- 238000001914 filtration Methods 0.000 claims description 17
- 230000009466 transformation Effects 0.000 claims description 11
- 230000011218 segmentation Effects 0.000 claims description 8
- 238000004364 calculation method Methods 0.000 description 15
- 239000011159 matrix material Substances 0.000 description 10
- 238000005259 measurement Methods 0.000 description 6
- 238000013016 damping Methods 0.000 description 4
- 230000004927 fusion Effects 0.000 description 4
- 238000012360 testing method Methods 0.000 description 4
- 238000012897 Levenberg–Marquardt algorithm Methods 0.000 description 3
- 238000010586 diagram Methods 0.000 description 3
- 238000002474 experimental method Methods 0.000 description 3
- 230000007246 mechanism Effects 0.000 description 3
- 238000006243 chemical reaction Methods 0.000 description 2
- 239000004020 conductor Substances 0.000 description 2
- 238000006073 displacement reaction Methods 0.000 description 2
- 230000005484 gravity Effects 0.000 description 2
- 230000001788 irregular Effects 0.000 description 2
- 239000002390 adhesive tape Substances 0.000 description 1
- 230000005540 biological transmission Effects 0.000 description 1
- 239000000919 ceramic Substances 0.000 description 1
- 230000000295 complement effect Effects 0.000 description 1
- 238000007796 conventional method Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000010292 electrical insulation Methods 0.000 description 1
- 239000000284 extract Substances 0.000 description 1
- 239000011521 glass Substances 0.000 description 1
- 238000003709 image segmentation Methods 0.000 description 1
- 238000003706 image smoothing Methods 0.000 description 1
- 238000009413 insulation Methods 0.000 description 1
- 238000013507 mapping Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000007781 pre-processing Methods 0.000 description 1
- 238000011897 real-time detection Methods 0.000 description 1
- 230000009467 reduction Effects 0.000 description 1
- 230000000087 stabilizing effect Effects 0.000 description 1
- 238000002945 steepest descent method Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T17/00—Three dimensional [3D] modelling, e.g. data description of 3D objects
- G06T17/20—Finite element generation, e.g. wire-frame surface description, tesselation
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01B—MEASURING LENGTH, THICKNESS OR SIMILAR LINEAR DIMENSIONS; MEASURING ANGLES; MEASURING AREAS; MEASURING IRREGULARITIES OF SURFACES OR CONTOURS
- G01B21/00—Measuring arrangements or details thereof, where the measuring technique is not covered by the other groups of this subclass, unspecified or not relevant
- G01B21/02—Measuring arrangements or details thereof, where the measuring technique is not covered by the other groups of this subclass, unspecified or not relevant for measuring length, width, or thickness
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/11—Region-based segmentation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/136—Segmentation; Edge detection involving thresholding
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/194—Segmentation; Edge detection involving foreground-background segmentation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/30—Determination of transform parameters for the alignment of images, i.e. image registration
- G06T7/33—Determination of transform parameters for the alignment of images, i.e. image registration using feature-based methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10028—Range image; Depth image; 3D point clouds
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Computer Graphics (AREA)
- Geometry (AREA)
- Software Systems (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses an insulator creepage distance detection method based on three-dimensional reconstruction, which comprises the following steps: for the point cloud data of the insulator to be detected, calculating a corresponding FPFH feature descriptor, finishing coarse registration of the point cloud data according to the obtained FPFH feature descriptor, finishing fine registration by using an LM algorithm, performing smoothing processing and point cloud curved surface reconstruction on the registered point cloud data, selecting a sampling plane based on the obtained three-dimensional reconstruction data, and calculating to obtain the creepage distance. Aiming at the problems that the manual detection method for the creepage distance of the insulator is complex, low in precision, incapable of being detected in an electrified mode and the like, the invention realizes the surface reconstruction and creepage distance detection of the insulator. The method has the advantages of high detection precision, convenient process, capability of detecting the insulator in an electrified mode, suitability for various insulators and the like, capability of greatly optimizing the detection flow of the creepage distance of the insulator and reducing manual operation.
Description
Technical Field
The invention relates to the technical field of insulator creepage distance detection, in particular to a three-dimensional reconstruction-based multi-type insulator creepage distance detection method and system.
Background
An insulator is a special insulating control, commonly used in overhead transmission lines, and is generally made of glass or ceramic. The insulator has two main functions: firstly, fixing and stabilizing a line to ensure the distance between current-carrying conductors; secondly, the creepage distance is increased, insulation is formed between the current carrying conductors and the ground, and accidents such as ground discharge and short circuit are prevented.
The insulator needs to have good structural strength and electrical insulation, so the creepage distance of the insulator is one of the important parameters for evaluating its quality. The creepage distance of an insulator refers to the shortest distance along the surface of the insulator between the locations where voltage is applied during normal operation of the insulator.
At present, creepage distance detection of an insulator mainly depends on manual work, and cannot be detected under the condition of electrification of the insulator. Meanwhile, the structure of the insulator is complex, particularly the umbrella skirt structure of the insulator is difficult to measure by a conventional method, and an adhesive tape detection method and a tape detection method are generally used for manual detection. The tape detection method is to use the tape to paste along the shortest distance between the voltage applying positions and the insulator surface, then tear the tape off to measure the distance, and the measurement process is complex and the precision cannot be ensured. The tape measure method is to measure the shortest distance along the surface of the insulator by using the tape measure, and the measurement accuracy is poor.
The manual detection needs to detect the insulator after the power is off, and is time-consuming and labor-consuming. Meanwhile, the measurement accuracy varies from person to person, and is difficult to ensure.
Patent document publication No. CN106910189a discloses an insulator creepage distance measuring system based on three-dimensional reconstruction, but there are problems such as: the three-dimensional reconstruction based on voxel data is carried out by adopting KinectFusion algorithm, not only color images are required to be acquired, but also the core of KinectFusion is linear ICP algorithm, the angle and displacement of transformation are directly calculated on the nearest point data of the two images, and the robustness of the algorithm is not high. And due to the defects of the reconstruction method, the reconstruction algorithm cannot reconstruct the too large insulator.
Disclosure of Invention
Aiming at the problems that the manual detection of the creepage distance of the insulator is time-consuming and labor-consuming, the electrified measurement cannot be performed, the measurement accuracy is poor and the like, the invention discloses a multi-type insulator creepage distance detection system based on three-dimensional reconstruction.
An insulator creepage distance detection method based on three-dimensional reconstruction comprises the following steps: for the point cloud data of the insulator to be detected, calculating a corresponding FPFH feature descriptor, finishing coarse registration of the point cloud data according to the obtained FPFH feature descriptor, finishing fine registration by using an LM algorithm, performing smoothing processing and point cloud curved surface reconstruction on the registered point cloud data, selecting a sampling plane based on the obtained three-dimensional reconstruction data, and calculating to obtain the creepage distance.
Preferably, the insulator creepage distance detection method based on three-dimensional reconstruction comprises the following steps:
(1) Acquiring point cloud data of an insulator to be detected;
(2) Calculating corresponding FPFH feature descriptors aiming at the obtained point cloud data;
(3) Coarse registration of the point cloud data by combining the obtained FPFH feature descriptors by using a SAC-IA algorithm;
(4) For the obtained rough registration point cloud data, finishing fine registration by using an LM algorithm (Levenberg-Marquardt);
(5) Smoothing the registered point cloud data and reconstructing a point cloud curved surface;
(6) Determining a sampling range for a point cloud curved surface meeting the requirements;
(7) Determining a sampling plane in a sampling range;
(8) And determining an effective point positioned on the sampling plane, and calculating the creepage distance.
According to the invention, the Azure Kinect sensor can be utilized to circumferentially collect depth information of different angles of the insulator to be detected, and the insulator to be detected can be fixed on the rotary round table during collection and can be matched with the rotary round table to rotate, so that the depth information of different angles of the insulator to be detected can be collected; and then converting the obtained depth information into point cloud information, and separating the point cloud information of the background and other objects by utilizing point cloud filtering and point cloud segmentation (namely, preprocessing operation of the point cloud data), so as to separate the point cloud data of the insulator to be detected independently. The Azure Kinect sensor is adopted, so that the data acquisition precision can be further improved, and the final calculation precision is further improved. In the actual detection, the sampling frequency of the sensor and the total sampling amount may be determined based on a predetermined experiment. For example, aiming at an insulator with a simpler structure, a small amount of depth images can be acquired at one angle, and the sampling amount at one angle can be increased as required. The total sampling amount and the sampling frequency can be increased according to the characteristics of the object to be detected so as to meet various precision requirements.
The point cloud filtering operation can remove irrelevant point clouds, reduce the point cloud density and interfere with the point cloud data, and is mainly completed by a direct filter, downsampling and removing outliers. Specifically, preferably, the point cloud filtering includes:
(1-1) removing points in the point cloud information, the coordinates of which are not within a set range, by using straight-through filtering;
(1-2) downsampling the point cloud information after the direct filtering treatment;
(1-3) removing outliers in the downsampled point cloud information.
In the invention, a pass through filter can be adopted to remove points with Z-axis (namely the distance between the insulator to be detected and the sensor) coordinates not in a set range, so as to remove background points at a far position and point cloud data at an origin point, namely (0, 0). Downsampling may be accomplished by VoxelGrid filters to create a three-dimensional voxel grid from the input point cloud data, and then within each voxel, approximating all points in the voxel with their centers of gravity. Through the downsampling operation, the density of the point cloud can be reduced, and the operand is reduced. Removing outliers may be accomplished by RadiusOutlierRemoval filters. Each point in the point cloud data has at least a set number of neighboring points within a certain range, otherwise it is filtered out. The filter can be used for removing error points generated when the Azure Kinect DK acquires depth information.
The FPFH feature descriptors describe k domain geometric attributes by parameterizing the spatial differences between query points and domain points and forming a multi-dimensional histogram. The Gao Weichao space in which the histogram is located provides a measurable information space for the feature representation, has invariance to the 6-dimensional pose of the point cloud corresponding surface, and is robust at different sampling densities or neighborhood noise levels.
The invention combines the SAC-IA algorithm with the FPFH characteristic descriptor to roughly align the point cloud data. The SAC-IA algorithm is an alignment algorithm based on the RANSAC algorithm that estimates a particular mathematical entity from a data set that may contain a large number of extra-office items. The algorithm can align and register two identical objects with different angles, and the work of roughly aligning point cloud data is completed.
Based on the point cloud data subjected to coarse registration, in order to perform fine registration on two point clouds with different angles, an optimized Levenberg-Marquardt algorithm is required to find an optimal value in an iterative mode. In order to increase accuracy of fine registration of two point clouds and efficiency of a registration process, the method sets a threshold value moderately in the initial process of fine registration of the two point clouds, and compares whether the error of transformation is smaller than the threshold value or not in each iteration set number. And if the maximum distance between the corresponding points is smaller than the threshold value, reducing the maximum distance between the corresponding points to further refine the point cloud registration accuracy, and accumulating the transformation matrix. Preferably, when the LM algorithm is used to accurately register the point cloud data, r times are iterated, comparing whether the error of transformation is smaller than a set threshold value, if so, reducing the set threshold value, and r=2 to 3.
In the invention, when the point cloud data is acquired, irregular data generated by errors can cause the phenomenon of unsmooth surface of the point cloud. In addition, after point cloud registration, point clouds of a certain area are easy to appear and are two overlapped curved surfaces. The problem can be solved by using a moving least squares (MLS, moving Least Squares) method to realize the estimation normal vector and the surface fitting to realize the point cloud smoothing. And (5) removing overlapped points and redundant points by utilizing point cloud smoothing, and increasing missing points. Preferably, the invention adopts a mobile least square method to carry out point cloud smoothing on the point cloud data after fine registration.
After the point cloud smoothing treatment, the invention reconstructs the point cloud surface through a poisson curved surface reconstruction algorithm. The poisson curved surface reconstruction algorithm is a triangular mesh reconstruction algorithm based on an implicit function, and an approximate curved surface is obtained by performing optimized interpolation processing on point cloud data.
In order to reduce the calculated amount and improve the calculation efficiency, the invention performs downsampling operation for each registered w groups of point cloud data after performing smoothing treatment and point cloud curved surface reconstruction on the registered point cloud data, wherein w=2-5. The point cloud registration calculation is carried out as the target point cloud after the fusion of the plurality of groups of point cloud data, so that the calculation efficiency is reduced, the point cloud density can be reduced through the point cloud downsampling, and the calculation amount is reduced. Through test, the downsampling operation is carried out once for every 3 groups of point cloud data registration, so that the point cloud registration efficiency can be effectively improved, and meanwhile, the point cloud data which are registered and fused at last can be ensured to meet the precision requirement.
In actual operation, after the point cloud curved surface model is obtained, judgment needs to be carried out to judge whether the obtained point cloud curved surface meets the precision requirement. If yes, the subsequent operation can be performed, and if not, the initial data acquisition step is returned, for example, the step (1) can be directly returned, the depth information of the insulator to be detected at different angles can be acquired again in the circumferential direction by using the depth sensor, and then the steps (1) to (5) such as the conversion and the pretreatment of the depth information are performed until the point cloud curved surface meeting the precision requirement is obtained. For the precision judgment, the automatic recognition can be realized by a computer, and the manual recognition can also be selected.
Selecting the nearest point between the voltage application parts of the obtained point cloud curved surface meeting the requirements by using a mouse, and determining a sampling range;
And determining the sampling plane by using the normal vector of any one of the two points and the vector between the two points, namely determining the sampling plane and the sampling range by using the normal vector of the selected point and the line segment between the selected points. Through the step, the adoption range is determined, the upper limit and the lower limit of the sampling range are determined by utilizing the two points, and the left and right boundaries of the sampling range are determined according to the size characteristics of the insulator to be detected (or the obtained point cloud curved surface size), so that a specific sampling plane is obtained. When the boundary is determined, the boundary can be automatically delimited by a computer or manually delimited. In addition, the sampling plane may be a curved surface including the whole point cloud or a curved surface including a part of the point cloud, but at least includes a complete part required to calculate the creepage distance, that is, at least includes a part on one side of the axis of the insulator to be detected.
After the sampling plane is obtained, various methods can be used to determine the effective point. In order to simplify the calculation process, the invention can preferably adopt an average sampling method, firstly uniformly determining a plurality of sampling points on a sampling plane, and then determining whether the sampling points are effective sampling points according to whether the sampling points are positioned on the determined point cloud curved surface. Specifically, when calculating the creepage distance:
Uniformly sampling in the determined sampling plane;
Determining effective sampling points according to whether the sampling points are positioned on the point cloud curved surface or not;
and sequentially calculating the distances of the two closest points according to the determined starting points to finally obtain the creepage distance.
Judging whether the sampling point is an effective sampling point or not, if the sampling point is not a point on the point cloud curved surface, judging that the sampling point is an ineffective sampling point, and eliminating the ineffective sampling point; and carrying out subsequent creepage distance calculation aiming at the collected effective sampling points.
As a specific preferred scheme, the insulator creepage distance detection method based on three-dimensional reconstruction comprises the following steps:
(1) Collecting depth information of an insulator on the rotary round table through an Azure Kinect sensor, and converting the depth information into point cloud data;
(2) Performing point cloud filtering and point cloud segmentation to remove the background and other objects, and independently separating out the point cloud data of the insulator;
(3) Calculating a Fast Point Feature Histogram (FPFH) feature descriptor;
(4) Roughly aligning point cloud data by using SAC-IA algorithm;
(5) Accurately matching point cloud data by using an optimized LM (Levenberg-Marquardt) algorithm;
(6) Smoothing the point clouds after registration and merging, and reconstructing a point cloud curved surface;
(7) If 3 groups of point cloud data are fused, downsampling the fused point cloud data, and reducing the point cloud density;
(8) If the reestablishing result of the insulator meets the requirement, selecting the nearest point between the voltage applying parts on the point cloud display interface by using a mouse;
(9) Determining a sampling plane and a sampling range by using the normal vector of the selected point and a line segment between the selected points;
(10) Calculating the creepage distance of the insulator by sampling the point cloud data of the surface of the insulator on the plane;
A three-dimensional reconstruction-based multi-type insulator creepage distance detection system comprises:
the depth sensor is positioned at one side of the insulator to be detected and used for acquiring depth images of different angles of the insulator to be detected;
the image processing unit is used for converting the depth image obtained by the depth sensor into point cloud information; then, performing point cloud filtering and point cloud segmentation on the obtained point cloud information, and independently separating out the point cloud data of the insulator; solving FPFH (field programmable gate array) feature descriptors of point cloud data of insulators to be detected; according to the obtained FPFH characteristic descriptor, coarse registration of point cloud data is completed, and fine registration is completed by using an LM algorithm; smoothing the registered point cloud data and reconstructing a point cloud curved surface;
And the sampling and calculating module is used for selecting a sampling plane based on the obtained three-dimensional reconstruction data and calculating to obtain the creepage distance.
The invention can rotate the depth sensor to collect depth images of the insulator to be detected at different angles, and can fix the insulator to be detected on a selectable platform, so that the depth sensor is not moved, and then the driving mechanism or man-made driving rotary platform is used for rotation, so as to obtain depth images of all angles.
Preferably, the depth sensor is an Azure Kinect sensor.
Preferably, the image processing unit includes:
the image conversion module is used for converting the depth image obtained by the depth sensor into point cloud information;
The image segmentation module is used for carrying out point cloud filtering and point cloud segmentation on the obtained point cloud information and independently separating out the point cloud data of the insulator;
the image registration module is used for solving FPFH feature descriptors of point cloud data of the insulator to be detected; according to the obtained FPFH characteristic descriptor, coarse registration of point cloud data is completed, and fine registration is completed by using an LM algorithm;
and the image reconstruction module is used for carrying out smoothing treatment and point cloud curved surface reconstruction on the registered point cloud data.
Preferably, the image registration module comprises a feature solving module, an image coarse registration module and an image fine registration module. The feature solving module is used for solving FPFH feature descriptors of the point cloud data of the insulators; the image coarse registration module completes coarse registration of the point cloud data according to the obtained FPFH feature descriptors; the image fine registration module is used for finishing fine registration by using an LM algorithm.
Preferably, the image reconstruction module comprises an image smoothing processing module and a point cloud curved surface reconstruction module.
Preferably, the sampling and calculating module includes:
The sampling module is used for determining a sampling range for the point cloud curved surface meeting the requirements; determining a sampling plane in a sampling range; then determining effective points on the sampling plane;
And the calculation module sequentially calculates the distance between the adjacent effective points to finally obtain the creepage distance.
One or more of the above modules may be an integrated unitary structure, such as an integrated control chip or a computer, and some modules may even be implemented manually by a human.
In the invention, a driving mechanism for driving the rotary round table to rotate and a controller for controlling the driving of the driving mechanism can be also arranged according to the requirement. Of course, the controller and any of the above modules may be integrated or controlled centrally.
In the invention, the creepage distance of the insulator to be detected can be obtained in real time by adopting a real-time detection and real-time calculation mode; the data information required by the acquisition can be arranged, and then the acquired data information is sequentially processed.
According to the invention, depth information of the insulator can be acquired through the Azure Kinect sensor, the depth information is converted into point cloud data, and then a plurality of pieces of point cloud data with different angles are acquired through the circular turntable for matching and fusion, so that a three-dimensional model of the insulator is reconstructed. And finally, calculating the shortest distance between the voltage applying positions along the surface of the insulator, thereby obtaining the creepage distance of the insulator. The whole system has the advantages of high detection precision, convenience in operation, wide application range, capability of detecting insulators in a charged mode, capability of measuring various types of insulators and the like.
The invention discloses a three-dimensional reconstruction-based insulator creepage distance detection method, which aims at the problems that an insulator creepage distance manual detection method is complex, low in precision, incapable of being detected in an electrified mode and the like, and realizes the surface reconstruction and creepage distance detection of an insulator. The method has the advantages of high detection precision, convenient process, capability of detecting the insulator in an electrified mode, suitability for various insulators and the like, capability of greatly optimizing the detection flow of the creepage distance of the insulator and reducing manual operation.
Drawings
Fig. 1 is a schematic flow chart of steps of an insulator creepage distance detection method based on three-dimensional reconstruction.
Fig. 2 is an illustration of insulator creepage distance detection according to the present invention, wherein the left image is point cloud data after insulator registration, and the right image is an effective point schematic diagram of the insulator creepage distance.
Detailed Description
In order to more particularly describe the present invention, the following detailed description of the technical scheme of the present invention is provided with reference to the accompanying drawings and the specific embodiments.
As shown in fig. 1, the hardware related to the invention mainly comprises a rotary round table for fixing an insulator to be detected; the Azure Kinect sensor is used for detecting the depth information of the insulator and is used for realizing image processing and creepage distance calculation. The detection method mainly comprises the following steps:
(1) The Azure Kinect sensor collects depth information.
The Azure Kinect sensor collects depth information of different angles of insulators on the rotary round table, and converts the obtained depth information into point cloud data. When converting the point cloud information data, an xy lookup table is pre-calculated according to parameters set by AzureKinect sensors and is used for storing x and y scale factors of each depth image pixel. When calculating the 3D X coordinates corresponding to the pixel points, multiplying the x scale factor of the pixel points of the depth image by the Z coordinates of the pixels (the Z axis value of the sensor coordinates, namely the distance between the sensor and the target, and the value of the pixel points on the depth image) to obtain the 3DX coordinates of the pixels. Similarly, the 3D Y coordinates can be calculated by multiplying the y scale factor.
In the step, depth information of different angles of the insulator on the rotary round table can be acquired in real time by adopting an Azure Kinect sensor, and the information acquisition is stopped after condition judgment is carried out until a point cloud curved surface (or a three-dimensional point cloud curved surface model) meeting the requirements is obtained. The depth image information with set quantity can be acquired, and then the obtained depth image information is processed step by step to obtain the three-dimensional point cloud curved surface model meeting the requirements.
(2) And (5) point cloud filtering and point cloud segmentation.
And removing the background and other objects by using point cloud filtering and point cloud segmentation, and independently separating out the point cloud data of the insulator. For the obtained point cloud data, irrelevant point clouds can be removed, the point cloud density and interference point cloud data can be reduced by utilizing point cloud filtering operation, and point cloud data corresponding to a target is obtained, wherein the point cloud filtering is mainly completed through a direct filter, downsampling and outlier removal. The pass-through filter PassThrough can remove points whose Z-axis coordinates are not within the set range, and is used to remove the background points at a far distance and point cloud data at the origin (typically, the center of the sensor lens), that is, (0, 0).
The downsampling is completed through VoxelGrid filters, a three-dimensional voxel grid is created by input point cloud data, then in each voxel, the gravity centers of all points in the voxel are used for approximately representing all points in the voxel, the density of the point cloud is reduced, and the operation amount is reduced.
RadiusOutlierRemoval filters can be implemented to remove outliers. Each point in the point cloud data has at least a set number of neighboring points within a certain range, otherwise it is filtered out. The filter can be used for removing error points generated when the Azure Kinect DK acquires depth information.
In order to obtain the point cloud data of the insulator, the interference of the desktop point cloud and the round platform point cloud needs to be removed. Thus, the point cloud on the plane and the cylinder point cloud are segmented out and removed by a random sample consensus (RANSAC) based algorithm. The RANSAC algorithm randomly extracts a sample subset from the samples, calculates model parameters for the subset by using a minimum variance estimation algorithm, calculates the deviation of the model parameters of all the samples, compares the deviation with the deviation by using a set threshold value, and the deviation is smaller than the threshold value, wherein the samples belong to interior points, and otherwise belong to exterior points.
(3) Fast Point Feature Histogram (FPFH) feature descriptors.
The PFH feature descriptors describe k domain geometric attributes of the point by parameterizing the spatial differences between the query points and the domain points and forming a multi-dimensional histogram. The Gao Weichao space in which the histogram is located provides a measurable information space for the feature representation, has invariance to the 6-dimensional pose of the point cloud corresponding surface, and is robust at different sampling densities or neighborhood noise levels. The fast point feature histogram (Fast Point Feature Histograms, FPFH) is a simplified form of PFH calculation. The idea is to compute a simplified point feature histogram (SIMPLIFIED POINT FEATURE HISTOGRAM, SPFH) for each of the k-neighbors of query point P q separately, and then reassign the nearest neighbors of each point, SPFH values will be used to weigh the values of FPFH, which step weights all SPFH into the final fast point feature histogram by the following formula:
Where the weight w k represents the distance between the query point P q and the neighbor point P k in the given metric space.
(4) And roughly registering the point cloud data.
And (3) roughly aligning point cloud data by combining the SAC-IA algorithm with the FPFH feature descriptor. The SAC-IA algorithm is an alignment algorithm based on the RANSAC algorithm, which estimates a specific mathematical entity from a data set that may contain a large number of extra-office items. The algorithm can align and register two identical objects with different angles, and the work of roughly aligning point cloud data is completed.
The SAC-IA algorithm takes FPFH feature descriptors of two point clouds with different angles as a data set, randomly selects some data points and uses the data points for random matching. And solving the rotation and displacement under the matching condition through SVD, testing all the remaining matching items, verifying whether the matching items meet the epipolar constraint obtained according to the matrix, and identifying all the matching items meeting the epipolar constraint (namely, the matching items with characteristic points very close to epipolar distances). These matches constitute a support set for the underlying matrix, the larger the support set the greater the likelihood that the calculated matrix is correct. Thus, the best results are obtained by repeatedly performing this process, leaving the matrix with the largest support set.
Specifically, the SAC-IA algorithm relies on the point feature histogram, so before executing this algorithm, the FPFH of the point cloud should be calculated, and the general idea of the algorithm is as follows:
And (4-1) selecting n sampling points from the point cloud P to be registered, wherein the distance between every two sampling points is larger than a preset minimum distance threshold d in order to ensure that the sampled points have different FPFH characteristics as much as possible.
(4-2) Searching one or more points with similar FPFH characteristics as the sampling points in the point cloud P in the target point cloud Q, and randomly selecting one point from the similar points as a one-to-one corresponding point of the point cloud P in the target point cloud Q.
And (4-3) calculating a rigid body transformation matrix between the corresponding points, and then judging the performance of the current registration transformation by solving the distance error sum function after the transformation of the corresponding points, so as to realize the coarse registration of the point cloud image.
(5) And precisely matching the point cloud data.
In order to perform fine registration on two point clouds with different angles, the source point cloud and the target point cloud which are subjected to rigid body transformation in coarse registration are required to be used as the source point cloud and the target point cloud in fine registration, and an optimal value is found in an iterative mode through an optimized Levenberg-Marquardt algorithm. In order to increase the accuracy of the fine registration of two point clouds and the efficiency of the registration process, the system sets an initial threshold (step threshold epsilon 2) to a moderate size in the fine registration point cloud process, and whether the error of the contrast transformation twice per iteration is smaller than the threshold or not. And if the maximum distance between the corresponding points is smaller than the threshold value, reducing the maximum distance between the corresponding points to further refine the point cloud registration accuracy, and accumulating the transformation matrix.
The Levenberg-Marquardt algorithm is an iterative algorithm for solving the least squares problem and can be seen as a combination of the steepest descent method and the Gauss-Newton method (GN method).
In the formula (1-1), P is an estimated quantity, P + is an optimal parameter, ε is a difference between an observed vector and an estimated observed vector, x is an observed vector, and f is a mapping of the parameter vector P to the estimated observed vectorIs a function of (2).
Taylor expansion in the field is shown in formula (1-2), wherein J is a Jacobian matrix
f(P+δP)≈f(P)+JδP (1-2)
Finding the step size δ P for each step of the iteration such that the equation (1-3) is minimized, it can be demonstrated that the optimal solution for the least squares exists when J δ P - ε is orthogonal to J, i.e., J T(JδP - ε) =0, resulting in an incremental normal equation for the GN method of equation (1-4):
||x-f(P+δP)||≈||x-f(P)-JδP||＝||ε-JδP|| (1-3)
JTJδP＝JTε (1-4)
the incremental normal equation for the LM algorithm introduces a damping term μ, where I is the identity matrix, as shown in equations (1-5).
(JTJ+μI)δP＝JTε (1-5)
If the covariance matrix Σ x of the observation vector x is obtained, a weighted normal equation is obtained as shown in the formula (1-6).
If the currently calculated delta P reduces the error, the update is accepted and the damping term mu is reduced, and if the current increment increases the function value, the damping term is increased and the normal equation is re-solved until an increment is calculated that reduces the function value. The LM algorithm will adjust the damping term μ every step to ensure error reduction.
LM algorithm termination condition: 1) Gradient magnitude, e.g., J T ε is below threshold ε 1; 2) The variation delta P of the step size is lower than the threshold epsilon 2; 3) The maximum number of iterations K max is reached.
(6) And (5) point cloud smoothing and point cloud curved surface reconstruction.
When the point cloud data is acquired, irregular data generated by errors can cause the phenomenon of unsmooth surface of the point cloud, and in addition, the point cloud of a certain area is easy to appear after the point cloud registration, and the point cloud is two overlapped curved surfaces. The problem can be solved by using a moving least squares (MLS, moving Least Squares) method to realize the estimation normal vector and the surface fitting to realize the point cloud smoothing. Through this step, an image as shown in the left diagram of fig. 2 is obtained.
Reconstructing the point cloud surface through a poisson curved surface reconstruction algorithm. The poisson curved surface reconstruction algorithm is a triangular mesh reconstruction algorithm based on an implicit function, and an approximate curved surface is obtained by performing optimized interpolation processing on point cloud data.
(7) The fusion point cloud is downsampled.
The point cloud registration calculation is carried out as the target point cloud after the fusion of the plurality of groups of point cloud data, so that the calculation efficiency is reduced, the point cloud density can be reduced through the point cloud downsampling, and the calculation amount is reduced. Tests show that the downsampling operation is carried out once for every 3 groups of point cloud data registration, so that the point cloud registration efficiency can be effectively improved, and meanwhile, the point cloud data which are registered and fused at last can meet the precision requirement.
(8) The mouse determines the sampling range.
And if the reconstruction result of the insulator meets the requirement, selecting the nearest points (M points and N points) between the voltage applying parts on the point cloud display interface by using a mouse, and determining the sampling range. And selecting the point cloud on the display interface by the mouse through the PCL callback function, and returning parameters of the point cloud. And (3) if the reconstruction result of the insulator does not meet the requirement, returning to the step (1) to acquire the image again or acquire the image in a complementary way.
(9) A sampling plane is determined.
The sampling plane is determined by two intersecting vectors, normal vector a= (x 1,y1,z1) of the first point (M point) and vector b= (x 2,y2,z2) between the two points. The normal vector of the sampling plane is obtained by calculating the outer product of the two vectors, and the formula is shown as (2-1).
The three-dimensional plane equation is shown in a formula (2-2), A, B and C are normal vectors of planes, and D is the distance from an origin to the planes.
Ax+By+Cz+D＝0 (2-2)
Determining the upper limit (a line passing through the M point and perpendicular to the axial direction of the insulator) and the lower limit (a line passing through the N point and perpendicular to the axial direction of the insulator) of a sampling plane by using point vectors corresponding to the two points (the M point and the N point) or manually selecting the two points; according to the size of the insulator image, the left boundary and the right boundary are manually selected, so that a rectangular sampling plane can be obtained. Of course, in the present invention and the present embodiment, the rectangle is not a necessary condition, as long as a plane containing the insulator creepage distance curve (the direction is determined by (2-1)) can be used as the sampling plane. The present embodiment is described taking a rectangular sampling plane as an example.
(10) And calculating the creepage distance.
And calculating the creepage distance of the insulator through the effective point of the creepage distance of the insulator on the sampling plane. Firstly, uniformly sampling on a sampling plane, judging whether each sampling point belongs to an insulator surface point or not, namely whether each sampling point belongs to a point on a constructed point cloud curved surface (point cloud surface), determining that the sampling point is an effective sampling point for a point with a judging result being yes, and judging whether the sampling point is not an effective sampling point one by one for a point with a judging result being no, so as to obtain the insulator surface point on the sampling plane, namely an effective point of an insulator creepage distance, as shown in a right diagram of fig. 2. And calculating the distance from one end point of the creepage distance curve of the insulator to the nearest point and the distance from the nearest point to the next nearest point, and continuously repeating the calculation operation until the other end point. The sum of the distances between all the nearest points is the creepage distance of the insulator.
Finally, for the three-dimensional reconstruction detection method, an accuracy comparison experiment is carried out with the traditional tape detection method and the traditional tape detection method, and the experiment is shown in table 1. As can be seen from Table 1, the method of the present invention can greatly save the detection time, greatly improve the measurement accuracy and obviously reduce the average error.
Table 1 comparison table of the performance of the same type of test mode according to the invention.
The method of the invention can rapidly and accurately detect the quality of the insulator, can be used for detecting the quality of the insulator which is just produced, can monitor the quality of the insulator which is being used, and further increases the safety of the insulator in the use process.
Finally, it should be noted that: the foregoing examples are provided for the purpose of illustration only and are not intended to be limiting, and although the invention has been described in detail with reference to examples, it will be understood by those skilled in the art that any modifications, equivalents, improvements, etc. that fall within the spirit and principles of the invention are intended to be included within the scope of the invention.
Claims (4)
1. The method for detecting the creepage distance of the insulator based on three-dimensional reconstruction is characterized by comprising the following steps of: calculating a corresponding FPFH feature descriptor aiming at point cloud data of an insulator to be detected, finishing rough registration of the point cloud data according to the obtained FPFH feature descriptor, finishing fine registration by using an LM algorithm, performing smoothing processing and point cloud curved surface reconstruction on the registered point cloud data, selecting a sampling plane based on the obtained three-dimensional reconstruction data, and calculating to obtain a creepage distance;
The method specifically comprises the following steps:
(1) Acquiring point cloud data of an insulator to be detected;
(2) Calculating corresponding FPFH feature descriptors aiming at the obtained point cloud data;
(3) Coarse registration of the point cloud data by combining the obtained FPFH feature descriptors by using a SAC-IA algorithm;
(4) Aiming at the obtained rough registration point cloud data, finishing fine registration by using an optimized LM algorithm;
(5) Smoothing the registered point cloud data and reconstructing a point cloud curved surface;
(6) Determining a sampling range for a point cloud curved surface meeting the requirements;
(7) Determining a sampling plane;
(8) Determining effective points on a sampling plane, and calculating a creepage distance;
Circumferentially acquiring depth information of different angles of an insulator to be detected by using an Azure Kinect sensor, converting the acquired depth information into point cloud information, and independently separating out the point cloud data of the insulator to be detected by using point cloud filtering and point cloud segmentation;
When the optimized LM algorithm is used for carrying out fine registration on point cloud data, comparing whether the error of transformation is smaller than a set threshold value or not r times per iteration, if so, reducing the maximum distance between corresponding points, wherein r=2-3;
after carrying out smoothing treatment and point cloud curved surface reconstruction on the registered point cloud data, carrying out downsampling operation on each registered w groups of point cloud data, wherein w=2-5;
Selecting the nearest point between the voltage application parts of the obtained point cloud curved surface meeting the requirements by using a mouse, and determining a sampling range; determining the sampling plane by using the normal vector of any one of the two points and the vector between the two points;
Uniformly sampling in the determined sampling plane;
Determining effective sampling points according to whether the sampling points are positioned on the point cloud curved surface or not;
And sequentially calculating the distances of the two closest points according to the determined starting points, and finally obtaining the creepage distance.
2. The three-dimensional reconstruction-based insulator creepage distance detection method of claim 1, wherein the point cloud filtering comprises:
(1) Removing points with coordinates not in a set range in the point cloud information by utilizing direct filtering;
(2) Downsampling the point cloud information after the direct filtering treatment;
(3) And removing outliers in the point cloud information after the downsampling process.
3. The three-dimensional reconstruction-based insulator creepage distance detection method of claim 1, wherein the point cloud smoothing is realized by adopting a mobile least square method, and a poisson curved surface reconstruction algorithm is adopted to reconstruct a point cloud curved surface.
4. An insulator creepage distance detection system based on three-dimensional reconstruction for realizing the detection method according to any one of claims 1 to 3, characterized by being positioned at one side of an insulator to be detected and being used for collecting depth images of different angles of the insulator to be detected;
the image processing unit is used for converting the depth image obtained by the depth sensor into point cloud information; then, performing point cloud filtering and point cloud segmentation on the obtained point cloud information, and independently separating out the point cloud data of the insulator; solving FPFH (field programmable gate array) feature descriptors of point cloud data of insulators to be detected; according to the obtained FPFH characteristic descriptor, coarse registration of point cloud data is completed, and fine registration is completed by using an LM algorithm; smoothing the registered point cloud data and reconstructing a point cloud curved surface;
and the sampling and calculating module is used for selecting a sampling plane based on the obtained three-dimensional reconstruction data and calculating to obtain the creepage distance.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110083707.3A CN112785711B (en) | 2021-01-21 | 2021-01-21 | Insulator creepage distance detection method and detection system based on three-dimensional reconstruction |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110083707.3A CN112785711B (en) | 2021-01-21 | 2021-01-21 | Insulator creepage distance detection method and detection system based on three-dimensional reconstruction |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN112785711A CN112785711A (en) | 2021-05-11 |
| CN112785711B true CN112785711B (en) | 2024-05-17 |
Family
ID=75758398
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110083707.3A Active CN112785711B (en) | 2021-01-21 | 2021-01-21 | Insulator creepage distance detection method and detection system based on three-dimensional reconstruction |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112785711B (en) |
Families Citing this family (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114264355B (en) * | 2021-11-18 | 2024-06-25 | 河南讯飞智元信息科技有限公司 | Weight detection method, device, electronic equipment and storage medium |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106296693A (en) * | 2016-08-12 | 2017-01-04 | 浙江工业大学 | Based on 3D point cloud FPFH feature real-time three-dimensional space-location method |
| CN106910189A (en) * | 2017-02-20 | 2017-06-30 | 广东电网有限责任公司惠州供电局 | A kind of insulator creep age distance measuring system and method based on three-dimensional reconstruction |
| CN107123161A (en) * | 2017-06-14 | 2017-09-01 | 西南交通大学 | A kind of the whole network three-dimensional rebuilding method of contact net zero based on NARF and FPFH |
| CN109920011A (en) * | 2019-05-16 | 2019-06-21 | 长沙智能驾驶研究院有限公司 | Outer ginseng scaling method, device and the equipment of laser radar and binocular camera |
-
2021
- 2021-01-21 CN CN202110083707.3A patent/CN112785711B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106296693A (en) * | 2016-08-12 | 2017-01-04 | 浙江工业大学 | Based on 3D point cloud FPFH feature real-time three-dimensional space-location method |
| CN106910189A (en) * | 2017-02-20 | 2017-06-30 | 广东电网有限责任公司惠州供电局 | A kind of insulator creep age distance measuring system and method based on three-dimensional reconstruction |
| CN107123161A (en) * | 2017-06-14 | 2017-09-01 | 西南交通大学 | A kind of the whole network three-dimensional rebuilding method of contact net zero based on NARF and FPFH |
| CN109920011A (en) * | 2019-05-16 | 2019-06-21 | 长沙智能驾驶研究院有限公司 | Outer ginseng scaling method, device and the equipment of laser radar and binocular camera |
Non-Patent Citations (3)
| Title |
|---|
| 低能见度条件下的航班延误以及恢复预测方法研究;苏靖晰;中国优秀硕士学位论文全文数据库 工程科技II辑(第11期);11-12 * |
| 基于SLAM三维重建的室内环境测绘研究;曾俊飞;硕士电子期刊信息科技辑;20200815(第8期);23-28 * |
| 基于三维重建的绝缘子爬电距离计算方法;彭刚等;电子测量技术;第40卷(第5期);159-162 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN112785711A (en) | 2021-05-11 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107063228B (en) | Target attitude calculation method based on binocular vision | |
| CN107798326B (en) | Contour vision detection method | |
| CN110443836A (en) | A kind of point cloud data autoegistration method and device based on plane characteristic | |
| CN107578400B (en) | BIM and three-dimensional point cloud fused contact network device parameter detection method | |
| CN108876744B (en) | Large-scale point cloud noise denoising method based on region segmentation | |
| CN109801333B (en) | Volume measurement method, device and system and computing equipment | |
| CN110807781B (en) | Point cloud simplifying method for retaining details and boundary characteristics | |
| CN108921164B (en) | Contact net locator gradient detection method based on three-dimensional point cloud segmentation | |
| CN113791400B (en) | Stair parameter autonomous detection method based on laser radar | |
| CN111598172B (en) | Dynamic target grabbing gesture rapid detection method based on heterogeneous depth network fusion | |
| CN111612850A (en) | Pig body size parameter measuring method based on point cloud | |
| CN108765495B (en) | Rapid calibration method and system based on binocular vision detection technology | |
| CN113532277B (en) | Method and system for detecting plate-shaped irregular curved surface workpiece | |
| CN110047133A (en) | A kind of train boundary extraction method towards point cloud data | |
| CN112241964B (en) | Light strip center extraction method for line structured light non-contact measurement | |
| CN112785711B (en) | Insulator creepage distance detection method and detection system based on three-dimensional reconstruction | |
| CN111968224A (en) | Ship 3D scanning point cloud data processing method | |
| CN107610174B (en) | Robust depth information-based plane detection method and system | |
| CN110517323A (en) | 3 D positioning system and method based on manipulator one camera multi-vision visual | |
| CN114998395A (en) | Effective embankment three-dimensional data change detection method and system | |
| Olson | Adaptive-scale filtering and feature detection using range data | |
| KR102171203B1 (en) | A method of matching a stereo image and an apparatus therefor | |
| Sun et al. | Triangular mesh construction based on point cloud matrix and edge feature extraction | |
| CN115186526A (en) | Ultra-high voltage transmission line discharge early warning method based on unmanned aerial vehicle monitoring | |
| Kawai et al. | Reconstruction of 3D objects by integration of multiple range data |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| CB03 | Change of inventor or designer information |
Inventor after: Chen Lingchao Inventor after: Hou Beiping Inventor after: Zhu Wen Inventor after: Jie Jing Inventor after: Yu Aihua Inventor after: Zhou Le Inventor after: Zhang Yong Inventor before: Hou Beiping Inventor before: Chen Lingchao Inventor before: Zhu Wen Inventor before: Jie Jing Inventor before: Yu Aihua Inventor before: Zhou Le Inventor before: Zhang Yong |
|
| CB03 | Change of inventor or designer information | ||
| CB02 | Change of applicant information |
Address after: 310023 No. 318 stay Road, Xihu District, Zhejiang, Hangzhou Applicant after: ZHEJIANG University OF SCIENCE AND TECHNOLOGY Address before: 310023 No. 318, Ho Ho Road, Hangzhou, Zhejiang Applicant before: ZHEJIANG University OF SCIENCE AND TECHNOLOGY |
|
| CB02 | Change of applicant information | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |