A method for identifying contact network positioning pillars based on 3D imaging technology

Abstract

The present invention discloses a contact network positioning pillar identification method based on 3D imaging technology, comprising the following steps: S1, coarsely locating the pillar area to obtain the pillar area; S2, performing secondary judgment on the pillar area; S3, outputting pillar-related attribute information based on the secondary judgment result. Pillar identification using three-dimensional point cloud features: Compared with two-dimensional images, the present invention uses three-dimensional point cloud overall judgment to be faster and more accurate. Adopting a secondary judgment method: The secondary judgment method provided by the present invention is a refined judgment based on the coarse positioning technology to avoid recognition errors caused by initial positioning abnormalities. Pillar attribute judgment: The present invention is an additional judgment performed on the premise of identifying the pillar, which is used to indirectly output the geometric relationship of the pillar.

Description

Technical Field

[0001] The present invention relates to the technical field of rail transit contact network pillar detection, and in particular to a contact network positioning pillar identification method based on 3D imaging technology. Background Art

[0002] Currently, the identification technology for contact network positioning pillars is primarily based on two-dimensional images combined with image recognition. Image recognition primarily employs two modes: 1. Using traditional image processing techniques to filter, binarize, and perform edge detection on the input image, combined with the geometric features of the pillars for identification. In practical applications, background interference in two-dimensional images is significant, and the contrast between the interference and the target is similar, making differentiation difficult. Furthermore, since the state of the target in a two-dimensional image varies at different distances, mostly reflected in its size, traditional image processing introduces instability in parameter settings, resulting in large recognition errors. 2. Using deep learning methods. However, deep learning relies heavily on samples, and the number and type of sample sets significantly influence the prediction results, leading to errors. These methods are not conducive to the accurate identification of contact network positioning pillars. Summary of the Invention

[0003] In view of the above-mentioned deficiencies in the prior art, the present invention provides a method for identifying contact network positioning pillars based on 3D imaging technology, which solves the problem that contact network positioning pillars cannot be accurately identified.

[0004] In order to achieve the above-mentioned object of the invention, the technical solution adopted by the present invention is: a method for identifying contact network positioning pillars based on 3D imaging technology, comprising the following steps:

[0005] S1. Coarsely locate the pillar area to obtain the pillar area, which includes the registration calculation of the key point feature descriptions of the template point cloud data and the point cloud data to be measured, and output the matching result;

[0006] S2. Perform a secondary judgment on the pillar area by calculating the geometric position relationship of the pillar area and comparing it with the geometric position relationship of the template point cloud data;

[0007] S3. Output the relevant attribute information of the positioning pillar according to the secondary judgment result.

[0008] Furthermore: the specific steps of step S1 are:

[0009] S11, sampling, key point extraction and key point feature description processing are performed on the template point cloud data containing only the pillar area;

[0010] The point cloud data to be tested is sampled, clustered, key points extracted and key point feature description processed in sequence;

[0011] S12. Perform registration calculation on the key point feature description of the template point cloud data and the key point feature description of the point cloud data to be measured, and output the matching result.

[0012] Further: the clustering includes:

[0013] Performing a downsampling operation on the point cloud data to be measured;

[0014] Use the DBSCAN density clustering method to divide the point cloud data to be tested into multiple clusters to obtain point cloud clusters;

[0015] The interference point cloud is filtered through the geometric properties of the circumscribed cube of the point cloud cluster, and the point cloud cluster of the pillar area is obtained according to the default parameters of the pillar area.

[0016] Furthermore, the key point extraction adopts Harris extraction method, the key point feature description adopts PFH, and the registration calculation adopts KD-TREE nearest neighbor query method.

[0017] Furthermore: the matching result includes:

[0018] The coordinate relationship of the point cloud data to be measured relative to the template point cloud data, wherein the coordinate relationship includes the angles of three coordinate axis planes and offsets in three directions.

[0019] Furthermore: the specific steps of step S2 are:

[0020] S21, input the matching result, perform coordinate transformation on the point cloud data to be measured, determine the ROI of the transformed data, and calculate the geometric position relationship of the associated area;

[0021] S22. Compare the ROI area and the geometric position relationship input in the template point cloud data with the geometric position relationship of the associated area, and output the distance information of the positioning pillar.

[0022] Furthermore, the ROI area in the template point cloud data is a circumscribed cube of the auxiliary structure of the pillar, and the geometric position relationship is the relationship between the auxiliary structure and the pillar body.

[0023] Furthermore: the step S21 is specifically as follows:

[0024] S211, input matching result, i.e. the angle θ between the three coordinate axis planes of the position change of the data to be measured relative to the template xoy ,θ xoz ,θ zoy and the offset x in three directions o 、y o 、z o ;

[0025] S212, through θ xoy ,θ xoz ,θ zoy 、x o 、y o 、z o Transform the point cloud data to be measured to the position of the template point cloud data, which is recorded as Cloud1;

[0026] S213 , using the circumscribed cube of the ROI to segment Cloud1 , recalculating the projection plane of the point cloud data within each cube, and calculating the relationship between the projection planes.

[0027] Furthermore, the step S22 is specifically as follows: comparing the relationship between the projection planes with the positional relationship in the template point cloud data; when the comparison result satisfies a certain threshold, the current point cloud data to be measured is a positioning pillar, and outputting the distance information of the positioning pillar.

[0028] Furthermore, the step S3 specifically includes: determining the geometric relationship of the pillars from the projection information calculated in the ROI cube of the pillar area, thereby determining whether the pillars have abnormal conditions such as deformation.

[0029] The beneficial effects of the present invention are:

[0030] 1. Pillar recognition using three-dimensional point cloud features: Compared with two-dimensional images, the present invention uses three-dimensional point clouds for overall judgment, which is faster and more accurate.

[0031] 2. Adopting a secondary judgment method: The secondary judgment method provided by the present invention performs a refined judgment based on the coarse positioning technology to avoid recognition errors caused by abnormal initial positioning.

[0032] 3. Pillar attribute judgment: The present invention performs additional judgment based on the premise of identifying the pillars, and is used to indirectly output the geometric relationship of the pillars. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] Figure 1 Flowchart of the present invention;

[0034] Figure 2 This is a flow chart of step S1 of the present invention;

[0035] Figure 3 This is a flow chart of step S2 of the present invention. DETAILED DESCRIPTION

[0036] The specific embodiments of the present invention are described below to facilitate understanding of the present invention by those skilled in the art. However, it should be clear that the present invention is not limited to the scope of the specific embodiments. For those skilled in the art, as long as various changes are within the spirit and scope of the present invention as defined and determined by the appended claims, these changes are obvious, and all inventions and creations utilizing the concepts of the present invention are protected.

[0037] like Figure 1 As shown, a method for identifying contact network positioning pillars based on 3D imaging technology includes the following steps:

[0038] S1. Coarsely locate the pillar area to obtain the pillar area, which includes the registration calculation of the key point feature description of the template point cloud data and the point cloud data to be measured, and output the matching result; the specific steps are as follows Figure 2 As shown,

[0039] S11. Sampling the template point cloud of the pillar coarse positioning, sampling and clustering the point cloud data to be measured, extracting key points from the sampling results, and performing feature description on the key points;

[0040] The template point cloud for pillar coarse positioning is a pre-prepared point cloud data containing only the pillar area. Its sampling, key point extraction, and key point feature description are consistent with the processing methods of the point cloud to be measured.

[0041] For data processing, the point cloud data is first downsampled. The DBSCAN density clustering method is then used to segment the point cloud data into multiple clusters. Finally, the interfering point clouds are filtered using the geometric properties of the point cloud cluster's bounding cube. The point cloud clusters for the pillar area are then obtained using the default parameters for the pillar area. This process is designed to reduce registration time and improve efficiency.

[0042] In the present invention, the Harris extraction method is used for key point extraction, the PFH (point feature histogram) is used for key point feature description, and the KD-TREE nearest neighbor query is used for feature registration.

[0043] S12. Perform registration calculation on the key point feature description of the template point cloud data and the key point feature description of the point cloud data to be measured, and output the matching result.

[0044] S2, by calculating the geometric position relationship of the pillar area and comparing it with the geometric position relationship of the template point cloud data, a secondary judgment is made on the pillar area; the specific steps are as follows: Figure 3 As shown,

[0045] S21, input the matching result, perform coordinate transformation on the data to be measured, determine the ROI of the transformed data, and calculate the geometric position relationship of the associated area;

[0046] S22. Compare the geometric position relationship between the ROI area and the geometric position relationship input in the template point cloud data and the geometric position relationship of the associated area, and output the distance information of the positioning pillar.

[0047] The ROI area in the template point cloud data is the circumscribed cube of the auxiliary structure of the pillar, and the geometric position relationship is the relationship between the auxiliary structure and the pillar body, such as angle, intersection, distance and other geometric information.

[0048] The main process of the secondary judgment is to first use the registration result obtained in step 1, that is, the angle θ between the three coordinate axis planes of the position change of the measured data relative to the template xoy ,θ xoz ,θ zoy and the offset x in three directions o 、y o 、z o , through the above six parameters, the point cloud data to be tested is transformed to the position of the template data, recorded as Cloud1. Then, the circumscribed cube of ROI is used to segment Cloud1, and the projection plane of the point cloud data in each cube is calculated again. The relationship between each projection plane is calculated and finally compared with the position relationship of the template. If a certain threshold is met, the current recognition data is considered as a pillar.

[0049] S3. Output pillar-related attribute information based on the secondary judgment result.

[0050] Projection information calculated from the ROI cube within the pillar area can indirectly determine the pillar's geometric relationship, thereby identifying any abnormalities such as deformation. For example, ideally, the pillar is perpendicular to the horizontal plane. However, if the calculated angle between the pillar plane and the horizontal plane is not 90 degrees, deformation is considered. 3D point cloud data can also be used to identify defects in the pillar's ancillary structures, such as damaged components, missing or loose key parts.

[0051] Pillar recognition using 3D point cloud features. Compared to 2D images, using 3D point clouds for overall judgment is faster and more accurate.

[0052] Secondary judgment method: The secondary judgment method provided by the present invention is a refined judgment based on the coarse positioning technology, which avoids the recognition error caused by the abnormal initial positioning.

[0053] Pillar attribute judgment is an additional judgment made on the premise of identifying the pillars, which is used to indirectly output the geometric relationship of the pillars.

Claims

1. A method for identifying contact network positioning pillars based on 3D imaging technology, characterized in that: The following steps are involved: S1. Coarsely locate the pillar area to obtain the pillar area, which includes the registration calculation of the key point feature descriptions of the template point cloud data and the point cloud data to be measured, and output the matching result; S2. Perform a secondary judgment on the pillar area by calculating the geometric position relationship of the pillar area and comparing it with the geometric position relationship of the template point cloud data; S3. Outputting the relevant attribute information of the positioning pillar according to the secondary judgment result; The specific steps of step S2 are: S21, input the matching result, perform coordinate transformation on the point cloud data to be measured, determine the ROI of the transformed data, and calculate the geometric position relationship of the associated area; S22. Compare the ROI area and the geometric position relationship input in the template point cloud data with the geometric position relationship of the associated area, and output the distance information of the positioning pillar.

2. The method for identifying contact network positioning pillars based on 3D imaging technology according to claim 1, characterized in that: The specific steps of step S1 are: S11, sampling, key point extraction and key point feature description processing are performed on the template point cloud data containing only the pillar area; The point cloud data to be tested is sampled, clustered, key points extracted and key point feature description processed in sequence; S12. Perform registration calculation on the key point feature description of the template point cloud data and the key point feature description of the point cloud data to be measured, and output the matching result.

3. The method for identifying contact network positioning pillars based on 3D imaging technology according to claim 2, characterized in that: The clustering includes: Performing a downsampling operation on the point cloud data to be measured; Use the DBSCAN density clustering method to divide the point cloud data to be tested into multiple clusters to obtain point cloud clusters; The interference point cloud is filtered through the geometric properties of the circumscribed cube of the point cloud cluster, and the point cloud cluster of the pillar area is obtained according to the default parameters of the pillar area.

4. The method for identifying contact network positioning pillars based on 3D imaging technology according to claim 2, characterized in that: The key point extraction adopts the Harris extraction method, the key point feature description adopts PFH, and the registration calculation adopts the KD-TREE nearest neighbor query method.

5. The method for identifying contact network positioning pillars based on 3D imaging technology according to claim 1, characterized in that: The matching results include: The coordinate relationship of the point cloud data to be measured relative to the template point cloud data, wherein the coordinate relationship includes the angles of three coordinate axis planes and offsets in three directions.

6. The method for identifying contact network positioning pillars based on 3D imaging technology according to claim 5, characterized in that: The ROI area in the template point cloud data is a circumscribed cube of the auxiliary structure of the pillar, and the geometric position relationship is the relationship between the auxiliary structure and the pillar body.

7. The method for identifying contact network positioning pillars based on 3D imaging technology according to claim 6, characterized in that: The step S21 is specifically as follows: S211. Input the matching result, i.e. the angle between the three coordinate axis planes of the position change of the point cloud data to be measured relative to the template. and the offset in three directions ; S212, pass Transform the point cloud data to be measured to the position of the template point cloud data, which is recorded as Cloud1; S213 , using the circumscribed cube of the ROI to segment Cloud1 , recalculating the projection plane of the point cloud data within each cube, and calculating the relationship between the projection planes.

8. The method for identifying contact network positioning pillars based on 3D imaging technology according to claim 6, characterized in that: The step S22 is specifically as follows: comparing the relationship between each projection plane with the positional relationship in the template point cloud data; when the comparison result meets a certain threshold, the current point cloud data to be measured is a positioning pillar, and outputting the distance information of the positioning pillar.

9. The method for identifying contact network positioning pillars based on 3D imaging technology according to claim 1, characterized in that: The step S3 specifically includes: determining the geometric relationship of the positioning pillar from the projection information calculated in the ROI cube of the positioning pillar, thereby determining whether the positioning pillar has abnormal deformation.

Images