Local feature-based point cloud segmentation method for narrow prism surface of miniature part

Abstract

The invention relates to the field of precise measurement of complex parts, in particular to a local feature-based point cloud segmentation method for a narrow prism surface of a miniature part. Comprising the following steps: acquiring original point cloud data of a part, and preprocessing the original point cloud data; a region growing algorithm based on seed point selection is used for the preprocessed data to obtain a large-region curved surface of the part except a narrow ridge surface; aligning the part model with the segmented large area to obtain a part narrow edge surface point cloud interval range adjacent to the large area; restoring the point cloud data in the narrow prismatic surface point cloud interval range, and calculating a three-dimensional point cloud curvature; constructing a reference point sequence based on the curvature of the three-dimensional point cloud, and judging whether the sequence can represent a narrow prism plane or not based on the curvature and the normal of the narrow prism; and a narrow prism surface is solved by using the finally obtained reference point, and point cloud segmentation is completed. According to the method, accurate registration and labeling of the point cloud data are not needed, the requirement for priori knowledge is reduced, the automation degree of point cloud segmentation is improved, and the method has high engineering application and popularization value.

Description

Technical Field

[0001] The invention relates to the field of complex parts precision measurement, in particular to a point cloud segmentation method at narrow edges of micro parts based on local features. Background Art

[0002] In actual engineering applications, there are many micro-complex structure parts that are core components, which have the characteristics of small size, precise structure, high technical requirements, and complex manufacturing process. Their key geometric features, such as the processing accuracy of the narrow edges of the parts, play a vital role in the overall system. The traditional geometric measurement method of mechanical parts cannot achieve the measurement of micro-complex structures. Digital measurement technology can discretize the surface geometry of the parts to be measured and convert it into a point cloud model. The point cloud processing algorithm can be used to analyze the dimensional error and shape error of the workpiece.

[0003] Point cloud segmentation is a key link in the point cloud data processing process, and the quality of the segmentation results has a great impact on the subsequent steps. Commonly used point cloud segmentation algorithms can be roughly divided into the following four types: edge-based, region-growing-based, model-based geometric elements, and deep learning-based point cloud segmentation algorithms. Edge-based and region-growing-based methods are suitable for instances where the geometric features or normal vector features of point cloud data have obvious differences and changes. The number of point clouds at the narrow edges of micro parts is small, and they are affected by edge measurement noise. The above segmentation methods cannot achieve automatic extraction; the point cloud segmentation method based on model geometric elements needs to align the point cloud data with the model first, and then calculate the distance from each point cloud data to each surface element for classification. Although this segmentation method has obvious basis, it has problems of low accuracy and under-segmentation when segmenting for parts with complex structures at narrow edges; the point cloud segmentation method based on deep learning requires the use of a large amount of labeled data and needs to be trained according to different scenarios. It is highly complex and takes a long time to calculate, and is not suitable for engineering applications. Summary of the invention

[0004] In view of the deficiencies in the prior art, the present invention provides a local feature-based point cloud segmentation method for narrow facets of micro-complex structure parts, which can automatically and accurately segment and extract point cloud data at the narrow facets of micro-parts.

[0005] The technical solution adopted by the present invention to achieve the above-mentioned purpose is:

[0006] A method for segmenting point clouds at narrow edges of micro parts based on local features comprises the following steps:

[0007] 1) Obtain the original point cloud data of the part and preprocess it;

[0008] 2) Using the region growing algorithm based on seed point selection on the preprocessed data, obtain the large area surface of the part except the narrow edge surface;

[0009] 3) Align the part model with the segmented large area to obtain the point cloud range of the narrow facets of the part adjacent to the large area;

[0010] 4) Restore the point cloud data within the narrow facet point cloud range and calculate the 3D point cloud curvature;

[0011] 5) Based on the curvature of the 3D point cloud, a reference point sequence is constructed, and based on the curvature and normal of the narrow facet, whether the sequence can represent the narrow facet plane is determined;

[0012] 6) Use the final reference points to find narrow edges and complete point cloud segmentation.

[0013] The step 1) is specifically as follows:

[0014] The original point cloud data is converted into voxel data through an octree-based voxelization process and then downsampled.

[0015] The step 2) is specifically as follows:

[0016] Calculate the curvature of the point cloud data of multiple points close to the center point in each voxel, sort the curvatures, select the point with the smallest curvature as the seed point, use the K-nearest neighbor or fixed radius search method to search its neighboring points, and use the growth criterion to compare the neighboring points one by one. If the growth criterion is met, the current neighboring point is added to the same block as the seed point, and if the curvature value of the neighboring point is less than the set curvature threshold, it is added to the current seed point queue; if the growth criterion is not met, continue to the next neighboring point judgment.

[0017] The use of the growth criteria for comparing neighboring points one by one is specifically: determining whether the normal angle between the seed point and the neighboring point is less than a set threshold.

[0018] The step 4) is specifically as follows:

[0019] By using the segmented large area surface and part model adjacent to the narrow facet, the normal of the plane where the narrow facet to be obtained is obtained, and the narrow facet point cloud interval is discretized according to the normal direction.

[0020] The step 5) is specifically as follows:

[0021] Calculate the curvature of the 3D point cloud in each discretization interval, select the point with the smallest curvature as the initial reference point ki, and put them into the initial reference point sequence K in sequence;

[0022] Using the large area surface and part model that have been segmented and are adjacent to the narrow facet, the plane normal threshold CT and curvature threshold ST of the narrow facet are set. If the plane normal and curvature of the initial reference point in the sequence K are both greater than the threshold, the reference point is removed from the sequence, otherwise it is retained.

[0023] The step 6) is specifically as follows:

[0024] Determine whether the number of remaining reference points after removing the sequence is less than 3. If so, it is determined that there are insufficient valid reference points on the narrow facet plane and the plane cannot be segmented and extracted. If the number of remaining reference points is greater than or equal to 3, select the 3 reference points with the smallest curvature as the final anchor points.

[0025] Assume the plane equation is ax+by+cz+d=0, the average normal direction of the three anchor points with the smallest curvature is taken as the normal direction of the split surface, the normal vector n=(a, b, c), and a set of parallel plane families is obtained;

[0026] Substitute the coordinates of the three anchor points into the plane equation to obtain the parameters d1, d2, and d3. Use the average value of d1, d2, and d3 as the d value of the plane equation to obtain the narrow facet to be determined.

[0027] The present invention has the following beneficial effects and advantages:

[0028] 1. Aiming at the problem that the narrow edge of the part has a complex structure, a limited number of point clouds, and is affected by edge measurement noise, the present invention effectively combines the local features of the narrow edge itself and the surrounding area to establish reference point selection with geometric constraints, which can achieve high-quality segmentation of the part point cloud and ensure the acquisition of high-precision results of subsequent dimensional errors and form and position errors;

[0029] 2. The present invention does not require precise registration and annotation of point cloud data, reduces the need for prior knowledge, improves the degree of automation of point cloud segmentation, and has strong engineering application and promotion value. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] FIG1 is a schematic diagram of a narrow facet of a complex part of the present invention and the corresponding triangulated mesh data;

[0031] Figure 2 Flowchart of the method for segmenting narrow facets of micro-complex structure parts designed for the present invention. DETAILED DESCRIPTION

[0032] The present invention is further described in detail below in conjunction with the accompanying drawings and embodiments.

[0033] Figure 1 is a schematic diagram of the narrow facets of a complex part of the present invention, a) is a model of the narrow facets of the complex part, b) is the triangulated data result obtained by packaging the point cloud file obtained by three-dimensional structured light scanning of the complex part, wherein plane 1 is the narrow facet to be measured. By comparing a) and b), it can be seen that the facet is too narrow, the edge measurement noise seriously affects the measurement accuracy, and the facet information cannot be obtained using existing segmentation methods.

[0034] Reference Figure 2 The method implementation steps of the present invention are as follows:

[0035] Step 1: Start the part point cloud segmentation by Figure 1b ) The original point cloud data obtained is used as input, and no alignment operation is required;

[0036] Step 2: Convert the input raw point cloud data into voxel data through an octree-based voxelization process and perform downsampling. Voxels are small cubes with a fixed length. Since point clouds are randomly distributed and have no regular topological structure, voxelizing the point cloud data into a regular grid structure can make the processing of point cloud data more convenient.

[0037] Step 3, use the regional growing algorithm based on seed point selection to obtain the large surface of the part. The specific method is to calculate the curvature of the point cloud data close to the center position in each voxel, select the point with the smallest curvature as the seed point after sorting, and use the K-nearest neighbor or fixed radius search method to search its neighboring points, and use the growth criterion (according to the angle between the seed point and the normal of the neighboring point is less than the set threshold) to compare them one by one. If the growth criterion is met, the current neighboring point is added to the same block as the seed point, and if the curvature value of the neighboring point is less than the set curvature threshold, it is added to the current seed point queue; if the growth criterion is not met, continue to the next neighboring point judgment.

[0038] Step 4: After step 3, the surface of most areas of the part can be automatically obtained. Only the narrow edges cannot be effectively segmented by step 3 due to the small amount of point cloud data and the influence of edge measurement noise. The part model is aligned with the segmented large area, and the point cloud interval range of the narrow edge of the part adjacent to the large area is calculated. This method is used to preliminarily filter out irrelevant discrete points.

[0039] Step 5, determine the range of the narrow facet point cloud interval, restore the point cloud data within the range, ensure sufficient feature point cloud data, calculate the three-dimensional point cloud curvature, implement the narrow facet point cloud segmentation method based on local features, select the narrow facet reference point sequence, and set the rules as follows:

[0040] Combined with the segmented adjacent faces and models, the normal of the narrow facet plane needs to be calculated, and the narrow facet point cloud interval is discretized in the normal direction. The size of the discrete range is set according to the actual geometric feature parameters of the part (set to 2mm*1mm*0.5mm in this example), and the data is not downsampled;

[0041] Step 6: Based on the local features of the narrow facet, determine whether the current reference point sequence can represent the narrow facet plane. The determination conditions are as follows:

[0042] Calculate the curvature of the 3D point cloud in each discretization interval, select the point with the smallest curvature as the initial reference point ki, and put it into the initial reference point queue K in turn.

[0043] Combined with the segmented adjacent faces and models, set the narrow facet plane normal threshold CT and curvature threshold ST. If the initial reference point in queue K is greater than the threshold, it is removed from the queue, otherwise it is retained.

[0044] Step 7: If the number of reference points remaining after step 6 is less than 3, it is determined that there are insufficient valid reference points on the narrow facet plane and the plane cannot be segmented and extracted; if the number of reference points remaining is greater than or equal to 3, select the 3 reference points with the smallest curvature as the final anchor points and execute step 8;

[0045] Step 8, assume that the plane equation is ax+by+cz+d=0, and the average normal direction of the three anchor points with the smallest curvature is taken as the normal direction of the segmentation surface, and the normal vector n=(a, b, c), so that a family of parallel planes can be obtained; then substitute the coordinates of the three anchor points into the plane equation to obtain d1, d2, and d3, and take the average value of d1, d2, and d3 as the d value of the plane equation. At this point, the narrow facet is obtained and the segmentation is completed.

Claims

1. A method for segmenting point clouds at narrow edges of micro parts based on local features, characterized in that: The following steps are involved: 1) Obtain the original point cloud data of the part and preprocess it; 2) Using the region growing algorithm based on seed point selection on the preprocessed data, obtain the large area surface of the part except the narrow edge surface; 3) Align the part model with the segmented large area to obtain the point cloud range of the narrow facets of the part adjacent to the large area; 4) Restore the point cloud data within the narrow facet point cloud range and calculate the 3D point cloud curvature; 5) Based on the curvature of the 3D point cloud, a reference point sequence is constructed, and based on the curvature and normal of the narrow facet, whether the sequence can represent the narrow facet plane is determined; 6) Use the final reference points to find narrow edges and complete point cloud segmentation.

2. According to the method of claim 1, the method is characterized in that: The step 1) is specifically as follows: The original point cloud data is converted into voxel data through an octree-based voxelization process and then downsampled.

3. The method for segmenting point clouds at narrow edges of micro-parts based on local features according to claim 1, characterized in that: The step 2) is specifically as follows: Calculate the curvature of the point cloud data of multiple points close to the center point in each voxel, sort the curvatures, select the point with the smallest curvature as the seed point, use the K-nearest neighbor or fixed radius search method to search its neighboring points, and use the growth criterion to compare the neighboring points one by one. If the growth criterion is met, the current neighboring point is added to the same block as the seed point, and if the curvature value of the neighboring point is less than the set curvature threshold, it is added to the current seed point queue; if the growth criterion is not met, continue to the next neighboring point judgment.

4. The method for segmenting point clouds at narrow edges of micro-parts based on local features according to claim 3, characterized in that: The use of the growth criteria for comparing neighboring points one by one is specifically: determining whether the normal angle between the seed point and the neighboring point is less than a set threshold.

5. The method for segmenting point clouds at narrow edges of micro-parts based on local features according to claim 1, characterized in that: The step 4) is specifically as follows: By using the segmented large area surface and part model adjacent to the narrow facet, the normal of the plane where the narrow facet to be obtained is obtained, and the narrow facet point cloud interval is discretized according to the normal direction.

6. The method for segmenting point clouds at narrow edges of micro-parts based on local features according to claim 1, characterized in that: The step 5) is specifically as follows: Calculate the curvature of the 3D point cloud of each discretization interval, select the point with the smallest curvature as the initial reference point ki, and put it into the initial reference point sequence K in sequence; Using the large area surface and part model that have been segmented and are adjacent to the narrow facet, the plane normal threshold CT and curvature threshold ST of the narrow facet are set. If the plane normal and curvature of the initial reference point in the sequence K are both greater than the threshold, the reference point is removed from the sequence, otherwise it is retained.

7. The method for segmenting point clouds at narrow edges of micro-parts based on local features according to claim 1, characterized in that: The step 6) is specifically as follows: Determine whether the number of remaining reference points after removing the sequence is less than 3. If so, it is determined that there are insufficient valid reference points on the narrow facet plane and the plane cannot be segmented and extracted. If the number of remaining reference points is greater than or equal to 3, select the 3 reference points with the smallest curvature as the final anchor points. Assume the plane equation is ax+by+cz+d=0, the average normal direction of the three anchor points with the smallest curvature is taken as the normal direction of the split surface, the normal vector n=(a, b, c), and a set of parallel plane families is obtained; Substitute the coordinates of the three anchor points into the plane equation to obtain the parameters d1, d2, and d3. Use the average value of d1, d2, and d3 as the d value of the plane equation to obtain the narrow facet to be determined.

Images