A Point Cloud Registration Method Combining Edge Features

Abstract

The present invention discloses a point cloud registration method combining edge features, which includes obtaining the point cloud of a template workpiece, searching for edge points and storing them in the template edge point cloud; obtaining the measured workpiece point cloud, searching for edge points and storing them in the measured edge point cloud; using the nearest neighbor search method to obtain the measured matching point set and the edge matching point set; constructing an objective function using the template workpiece point cloud, the measured matching point set, the template edge point cloud, and the edge matching point set; and using an optimization method to iteratively solve the objective function to obtain the rotation matrix R and the translation matrix t, thereby completing the registration between the template workpiece point cloud and the measured workpiece point cloud. This method aims to solve the registration problem in the case where the surface topography of the workpiece has a small undulation and the texture is not rich.

Description

Technical Field

[0001] The present invention relates to the field of point cloud registration, and specifically relates to a point cloud registration method combining edge features. Background Art

[0002] Point Cloud Registration refers to inputting two point clouds Ps, (s represents source) and Pt, (t represents target), and outputting a rotation and translation relationship to make the coincidence degree of Ps and Pt as high as possible. Currently, the traditional point cloud registration method is the ICP method (Iterative Closest Point). The core idea of the ICP method is to solve the transformation based on the nearest points of the current Ps in Pt and repeatedly iterate to obtain the optimal solution. In scenarios where the surface features of the workpiece to be measured are relatively rich, the registration effect of this method is accurate. However, for some workpieces with small surface topography undulations and scarce textures, the surface features of the collected point cloud are few, such as plate-shaped flat workpieces. At this time, using the ICP method will converge to a local minimum, resulting in only partial coincidence of the point cloud in the registration result, without achieving overall complete alignment, and there are deviations at the edges of the features (such as Figure 1 、 Figure 2 ), and the accuracy of the registration result is low, which cannot be applied to the fields of high-precision machining and assembly. Summary of the Invention

[0003] In order to solve the above technical problems, the present invention provides a point cloud registration method combining edge features, aiming to solve the problem that the existing ICP method has inaccurate registration results in the case of small surface topography undulations and scarce textures of the workpiece. For such workpieces, adding edge feature constraints of the workpiece can obtain more accurate registration results.

[0004] Therefore, the technical solution of the present invention is as follows:

[0005] A point cloud registration method combining edge features, which acquires a reference workpiece point cloud, frames and selects the point cloud of the region of interest therein and stores it as a template workpiece point cloud, searches for the edge points in this point cloud, and stores them in the template edge point cloud;

[0006] The following steps are used for point cloud registration:

[0007] 1) Acquire the actual scene point cloud, perform preprocessing, store the preprocessed point cloud as the measured workpiece point cloud, search for the edge points in this point cloud, and store them in the measured edge point cloud;

[0008] The measured workpiece point cloud contains the workpiece point cloud of the same model as the reference workpiece;

[0009] For the three-dimensional points in the template workpiece point cloud, use the nearest neighbor search method to find the point in the measured workpiece point cloud that is closest to it, and store the found point in the measured matching point set;

[0010] For the three-dimensional points in the template edge point cloud, use the nearest neighbor search method to find the point in the measured edge point cloud that is closest to it, and store all the found points in the edge matching point set;

[0011] 2) Construct the objective function E(R, t):

[0012]

[0013] where p i represents a point in the template workpiece point cloud, q i represents a point in the measured matching point set, k i represents the normal vector of q i , n represents the number of points in the template workpiece point cloud; h j represents a point in the template edge point cloud, g j represents a point in the edge matching point set, m represents the number of points in the template edge point cloud; λ is a preset weight value;

[0014] Use the optimization method to iteratively solve the objective function E(R, t) to obtain the rotation matrix R and the translation matrix t, and complete the registration between the template workpiece point cloud and the measured workpiece point cloud.

[0015] Preferably, the method for searching for edge points in the point cloud is:

[0016] In the point cloud, use the normal-based boundary extraction method to obtain edge points.

[0017] Preferably, while obtaining the reference workpiece point cloud and the actual scene point cloud, a two-dimensional image is also collected at the same position; there is a one-to-one correspondence between the pixel points in the two-dimensional image and the respective three-dimensional points in the point cloud;

[0018] Record the two-dimensional image corresponding to the reference workpiece point cloud as the reference two-dimensional image; record the two-dimensional image corresponding to the point cloud in the actual scene as the measured two-dimensional image;

[0019] The method for searching for edge points in the template workpiece point cloud is:

[0020] Perform edge extraction in the reference two-dimensional image to obtain the pixel coordinates of each edge point, and find the three-dimensional points corresponding to each pixel coordinate in the template workpiece point cloud, which are recorded as edge points and stored in the template edge point cloud;

[0021] The method for searching for edge points in the measured workpiece point cloud is:

[0022] Edge extraction is performed on the measured two-dimensional image to obtain the pixel coordinates of each edge point, and the three-dimensional points corresponding to each pixel coordinate in the measured workpiece point cloud are found, denoted as edge points, and stored in the measured edge point cloud.

[0023] Preferably, while obtaining the reference workpiece point cloud and the actual scene point cloud, a two-dimensional image is also collected at the same position; there is a one-to-one correspondence between the pixel points in the two-dimensional image and the three-dimensional points in the point cloud.

[0024] The two-dimensional image corresponding to the reference workpiece point cloud is denoted as the reference two-dimensional image, and the two-dimensional image corresponding to the point cloud in the actual scene is denoted as the measured two-dimensional image.

[0025] The template workpiece point cloud is denoted as point cloud I, and the pixel points corresponding to each point in point cloud I in the reference two-dimensional image are stored in point set I.

[0026] Use the following steps to search for edge points in point cloud I:

[0027] ① Respectively take each pixel point in point set I as the center point, search for the pixel points in the n×n neighborhood around the center point, store the three-dimensional points corresponding to the found pixel points in point cloud I as neighborhood points, and denote the point corresponding to the center point in point cloud I as the reference point.

[0028] ② Calculate the distance between each neighborhood point and the reference point, count the total number B of points with a distance value less than threshold I, and the total number C of points with a distance value greater than or equal to 2 - 4 times threshold I. Judge whether B and C are within the preset interval. If so, store the reference point as an edge point; otherwise, do not store it.

[0029] In step 1), denote the measured workpiece point cloud as the new point cloud I, and store the pixel points corresponding to each point in the new point cloud I in the measured two-dimensional image in the new point set I; use the new point cloud I and the new point set I to perform steps ① and ② to search for edge points.

[0030] Preferably, threshold I is d is the average point spacing of the point cloud; n = 3, 5, 7, or 9; the preset interval is:

[0031] Furthermore, while obtaining the reference workpiece point cloud and the actual scene point cloud, a two-dimensional image is also collected at the same position; there is a one-to-one correspondence between the pixel points in the two-dimensional image and the three-dimensional points in the point cloud.

[0032] Denote the two-dimensional image corresponding to the reference workpiece point cloud as the reference two-dimensional image; denote the two-dimensional image corresponding to the point cloud in the actual scene as the measured two-dimensional image.

[0033] In step 1), perform the following preprocessing on the actual scene point cloud to screen out the measured workpiece point cloud from the actual scene point cloud:

[0034] Find each two-dimensional pixel point corresponding to the template workpiece point cloud in the reference two-dimensional image, and denote the image area formed by all the found pixel points as Area One; manually set a selection box in the reference two-dimensional image, where the number of pixel points within the selection box is greater than that of Area One and completely covers Area One; save the image area within the selection box as a feature map, and select at least four points on the feature map as feature points;

[0035] Taking each feature point as the center, frame a local small area in the reference two-dimensional image. In step 1), use each local small area to perform template matching in the measured two-dimensional image, denote the matched area as Area Two, and denote the center point of Area Two as the adaptation point;

[0036] Solve the homography matrix using the feature points and the adaptation points;

[0037] Then use the homography matrix to convert each single two-dimensional pixel point on the feature map to the measured two-dimensional image respectively, obtain its pixel coordinates in the measured two-dimensional image, and store the three-dimensional point corresponding to the pixel coordinates in the point cloud of the actual scene into the measured workpiece point cloud; traverse each two-dimensional pixel point on the feature map, and store the corresponding three-dimensional points into the measured workpiece point cloud in the same way.

[0038] Preferably, the feature points are evenly distributed in the feature map.

[0039] Preferably, when using the optimization method to iteratively solve the objective function E(R, t), the initial values of the rotation matrix R and the translation matrix t are obtained through the SVD decomposition of the homography matrix.

[0040] Preferably, the nearest neighbor search method in step 1) includes the kd-tree search method or the octree search method;

[0041] The optimization method in step 2) is the least squares method, the Gauss-Newton method or the Levenberg-Marquardt method;

[0042] Among them,

[0043] Furthermore, when using the optimization method to iteratively solve the objective function E(R, t), the initial values of the rotation matrix R and the translation matrix t are obtained through coarse registration, or the initial values of the rotation matrix R and the translation matrix t are set as the identity matrix.

[0044] Compared with the prior art, the method provided by this application has the following advantages:

[0045] (1) The traditional ICP method only constrains the fitting of the overall surface topography point cloud of the workpiece, and the registration result can only ensure the positioning accuracy of the overall contour of the workpiece, and it is easy to have the problem of incorrect local feature registration. The method of the present invention takes into account both the overall surface topography information and the edge information of local features of the workpiece, and can obtain a more accurate registration result.

[0046] (2) In the process of extracting edge points, a method combining two-dimensional images and three-dimensional point clouds is adopted. Through conditional constraints, it is ensured that there are continuous other edge points around the selected edge points and there are large faults in space, which conforms to the edge characteristics, effectively suppresses the interference of noise points, and extracts more accurate edges.

[0047] (3) By combining template matching and homography matrix calculation, when selecting the workpiece area point cloud in the actual scene point cloud, an effective registration area can be quickly obtained, reducing the number of point clouds participating in the calculation and improving the calculation speed. Description of the Drawings

[0048] Figure 1 is a schematic diagram of the point cloud registration effect of the traditional IPC method;

[0049] Figure 2 is Figure 1 a partial enlarged view of the circular hole area in;

[0050] Figure 3 is a schematic diagram of the registration effect of the method of the present invention in the specific embodiment;

[0051] Figure 4 is Figure 3 a partial enlarged view of the circular hole area in;

[0052] Figure 5 is a schematic diagram of obtaining a feature map and feature points in the reference two-dimensional image in the specific embodiment. Specific Embodiment

[0053] The technical solution of the present invention will be described in detail below in conjunction with the drawings and specific embodiments.

[0054] A point cloud registration method combining edge features, obtaining a reference workpiece point cloud, selecting the point cloud of the interested area therein and storing it as a template workpiece point cloud, searching for edge points in the point cloud, and storing them in the template edge point cloud;

[0055] Using the following steps for point cloud registration:

[0056] 1) Obtain the actual scene point cloud, perform preprocessing, store the preprocessed point cloud as the measured workpiece point cloud, search for edge points in the point cloud, and store them in the measured edge point cloud;

[0057] The measured workpiece point cloud contains the workpiece point cloud of the same model as the reference workpiece;

[0058] For the three-dimensional points in the template workpiece point cloud, use the nearest neighbor search method to find the point in the measured workpiece point cloud that is closest to it in distance, and store the found point in the measured matching point set;

[0059] For the three-dimensional points in the template edge point cloud, use the nearest neighbor search method to find the point in the measured edge point cloud that is closest to it in distance, and store all the found points in the edge matching point set;

[0060] Among them, the nearest neighbor search method can adopt the kd-tree search method or the octree search method.

[0061] 2) Construct the objective function E(R, t):

[0062]

[0063] Among them, p i represents a point in the template workpiece point cloud, q i represents a point in the measured matching point set, k i represents the normal vector of q i , n represents the number of points in the template workpiece point cloud; h j represents a point in the template edge point cloud, g j represents a point in the edge matching point set, m represents the number of points in the template edge point cloud; λ is a preset weight value;

[0064] Use the optimization method to iteratively solve the objective function E(R, t) to obtain the rotation matrix R and the translation matrix t, and complete the registration between the template workpiece point cloud and the measured workpiece point cloud.

[0065] Specifically, the optimization method is the least squares method, the Gauss-Newton method or the Levenberg-Marquardt method; the initial values of the rotation matrix R and the translation matrix t are obtained through rough registration, or the initial values of the rotation matrix R and the translation matrix t are set to the identity matrix.

[0066] As a preferred setting method,

[0067] Specifically, for the method of searching for edge points in the point cloud, the following three methods can be selected:

[0068] Method 1:

[0069] In the point cloud, use the normal-based boundary extraction method to obtain edge points.

[0070] That is, use the normal-based boundary extraction method to search for edge points in the template workpiece point cloud and store them in the template edge point cloud.

[0071] Use the normal-based boundary extraction method to search for edge points in the measured workpiece point cloud and store them in the measured edge point cloud.

[0072] Method 2:

[0073] While obtaining the reference workpiece point cloud and the actual scene point cloud, also collect a 2D image at the same position; there is a one-to-one correspondence between the pixel points in the 2D image and the respective 3D points in the point cloud.

[0074] Denote the 2D image corresponding to the reference workpiece point cloud as the reference 2D image; denote the 2D image corresponding to the point cloud in the actual scene as the measured 2D image.

[0075] The method for searching for edge points in the template workpiece point cloud is as follows:

[0076] Perform edge extraction (Canny, sobel) on the reference 2D image to obtain the pixel coordinates of each edge point, and find the corresponding 3D points in the template workpiece point cloud for each pixel coordinate, denote them as edge points, and store them in the template edge point cloud.

[0077] The method for searching for edge points in the measured workpiece point cloud is as follows:

[0078] Perform edge extraction on the measured 2D image to obtain the pixel coordinates of each edge point, and find the corresponding 3D points in the measured workpiece point cloud for each pixel coordinate, denote them as edge points, and store them in the measured edge point cloud.

[0079] Method 3:

[0080] While obtaining the reference workpiece point cloud and the actual scene point cloud, also collect a 2D image at the same position; there is a one-to-one correspondence between the pixel points in the 2D image and the respective 3D points in the point cloud.

[0081] Denote the 2D image corresponding to the reference workpiece point cloud as the reference 2D image, and the 2D image corresponding to the point cloud in the actual scene as the measured 2D image.

[0082] Denote the template workpiece point cloud as point cloud I, and store the pixel points corresponding to each point in point cloud I in the reference 2D image into point set I.

[0083] Use the following steps to search for edge points in point cloud I:

[0084] ① For each pixel point in the point set I as the center point, search for the pixel points within the n×n neighborhood around the center point, store the three-dimensional points corresponding to the found pixel points in the point cloud I as neighborhood points, and record the point corresponding to the center point in the point cloud I as the reference point;

[0085] ② Calculate the distances between each neighborhood point and the reference point, count the total number B of points with distance values less than threshold I, and the total number C of points with distance values greater than or equal to 2 - 4 times threshold I. Determine whether B and C are within the preset interval. If so, store the reference point as an edge point; otherwise, do not store it;

[0086] In step 1), record the measured workpiece point cloud as the new point cloud I, store the pixel points corresponding to each point in the new point cloud I in the measured two-dimensional image into the new point set I; use the new point cloud I and the new point set I to perform steps ① and ② to search for edge points.

[0087] Among them, threshold I is (the maximum distance between the corner point and the center point in the n×n neighborhood), d is the average point spacing of the point cloud;

[0088] The calculation method of the average point spacing of the point cloud is as follows: build a k-d tree for the point cloud; calculate the average Euclidean distance between each point and its corresponding k neighboring points; take the average of the average Euclidean distances corresponding to all points as the average point spacing of the point cloud.

[0089] n = 3, 5, 7 or 9; the preset interval is:

[0090] Next, take the Figure 3 workpiece in as an example for an exemplary elaboration:

[0091] A point cloud registration method combining edge features, obtain the reference workpiece point cloud, frame and select the point cloud of the region of interest therein (remove the background point cloud and only retain the workpiece point cloud) and store it as the template workpiece point cloud, search for the edge points in this point cloud, and store them in the template edge point cloud;

[0092] Use the following steps for point cloud registration:

[0093] 1) Obtain the actual scene point cloud, perform preprocessing, store the preprocessed point cloud as the measured workpiece point cloud, search for the edge points in this point cloud, and store them in the measured edge point cloud;

[0094] The measured workpiece point cloud contains the workpiece point cloud of the same model as the reference workpiece; the measured workpiece point cloud not only contains the workpiece point cloud but also part of the point cloud in the area around the workpiece;

[0095] In this embodiment, perform the following preprocessing on the actual scene point cloud to screen out the measured workpiece point cloud from the actual scene point cloud:

[0096] While acquiring the point cloud of the reference workpiece and the point cloud of the actual scene, a two-dimensional image is also acquired at the same position; there is a one-to-one correspondence between the pixel points in the two-dimensional image and the three-dimensional points in the point cloud;

[0097] The two-dimensional image corresponding to the reference workpiece point cloud is denoted as the reference two-dimensional image; the two-dimensional image corresponding to the point cloud in the actual scene is denoted as the measured two-dimensional image;

[0098] As Figure 5 shown, find each two-dimensional pixel point corresponding to the template workpiece point cloud in the reference two-dimensional image, and denote the image area formed by all the found pixel points as Area One; manually set a selection box in the reference two-dimensional image, the number of pixel points within the selection box is greater than Area One and completely covers Area One; save the image area within the selection box as a feature map, and select at least four points on the feature map as feature points;

[0099] That is: the reference two-dimensional image contains the feature map, and the feature map contains Area One.

[0100] In order to obtain a more accurate homography matrix, the feature points are evenly distributed in the feature map. In this embodiment, for the convenience of searching, as Figure 5 shown, the feature points are the four corner points of the feature map.

[0101] Taking each feature point as the center, frame a local small area in the reference two-dimensional image. In step 1), use each local small area to perform template matching in the measured two-dimensional image, denote the matched area as Area Two, and denote the center point of Area Two as the adaptation point;

[0102] Solve the homography matrix using multiple feature points and multiple adaptation points;

[0103] Then use the homography matrix to convert the individual two-dimensional pixel points on the feature map to the measured two-dimensional image respectively, obtain their pixel coordinates in the measured two-dimensional image, and store the three-dimensional points corresponding to the pixel coordinates in the point cloud of the actual scene into the measured workpiece point cloud; traverse each two-dimensional pixel point on the feature map, and store the corresponding three-dimensional points into the measured workpiece point cloud in the same way.

[0104] Next, perform the following processing:

[0105] For the three-dimensional points in the template workpiece point cloud, use the kd-tree search method to find the point in the measured workpiece point cloud that is closest to it in distance, and store the found point into the measured matching point set;

[0106] For the three-dimensional points in the template edge point cloud, use the kd-tree search method to find the point in the measured edge point cloud that is closest to it in distance, and store all the found points into the edge matching point set;

[0107] 2) Construct the objective function E(R, t):

[0108]

[0109] where p i represents a point in the point cloud of the template workpiece, q i represents a point in the measured matching point set, k i represents the normal vector of q i , n represents the number of points in the point cloud of the template workpiece; h j represents a point in the point cloud of the template edge, g j represents a point in the edge matching point set, m represents the number of points in the point cloud of the template edge; λ is a preset weight value;

[0110] Use the least squares method to iteratively solve the objective function E(R, t) to obtain the rotation matrix R and the translation matrix t, and complete the registration between the point cloud of the template workpiece and the point cloud of the measured workpiece.

[0111] Among them, the initial values of the rotation matrix R and the translation matrix t are obtained by SVD decomposition of the homography matrix.

[0112] The point cloud registration effect processed by this method, as Figure 3 , 4 shown, this method can not only align the whole workpiece, but also achieve alignment in details (especially at the edges), and the positioning is more accurate; in contrast, the processing effect of the traditional ICP method is as Figure 1 , 2 shown, there are large registration errors at the edges of the L-shaped hole and the round hole, and only the surface topography of the point cloud is fitted, and it is difficult to achieve edge alignment.

[0113] In order to more intuitively display the error comparison between this method and the traditional ICP method, after registering with this method and the traditional ICP method respectively, calculate the deviation between the center coordinates of the round hole in the point cloud of the template workpiece after point cloud registration and the center coordinates of the round hole in the point cloud of the measured workpiece, that is: in Figure 1 , Figure 3 calculate the coordinate deviation and Euclidean distance between the two center coordinates respectively, and the specific data are as follows:

[0114] Registration method X-axis deviation / mm Y-axis deviation / mm Z-axis deviation / mm Euclidean distance Traditional ICP method 0.871 -0.270 0.010 0.912 Method of the present invention -0.128 -0.045 0.015 0.137

[0115] From the data, it can be seen that after registration by this method, the deviation value in the round hole area is less than 0.2 mm, and the alignment degree of the point cloud in the template workpiece and the point cloud in the measured workpiece is better, which is suitable for point cloud processing in precision machining manufacturing.

[0116] The foregoing description of the specific exemplary embodiments of the present invention has been presented for purposes of illustration and description. The foregoing description is not intended to be exhaustive or to limit the invention to the precise forms disclosed, and obviously many modifications and variations are possible in light of the above teachings. The exemplary embodiments were chosen and described in order to explain specific principles of the invention and its practical application so as to enable others skilled in the art to make and utilize the various exemplary embodiments of the invention and their various alternatives and modifications. The scope of the invention is intended to be defined by the appended claims and their equivalents.

Claims

1. A point cloud registration method combining edge features, which acquires the point cloud of a reference workpiece, selects and stores the point cloud of the region of interest therein as the point cloud of the template workpiece, searches for the edge points in the point cloud of the template workpiece, and stores them into the template edge point cloud; Characterized in that, Use the following steps to perform point cloud registration: 1) Acquire the point cloud of the actual scene, perform preprocessing, store the preprocessed point cloud as the measured workpiece point cloud, search for the edge points in the measured workpiece point cloud, and store them into the measured edge point cloud; The measured workpiece point cloud contains the point cloud of the workpiece with the same model as the reference workpiece; For the three-dimensional points in the point cloud of the template workpiece, use the nearest neighbor search method to find the point in the measured workpiece point cloud that is closest to it in distance, and store the found point into the measured matching point set; For the three-dimensional points in the template edge point cloud, use the nearest neighbor search method to find the point in the measured edge point cloud that is closest to it in distance, and store all the found points into the edge matching point set; 2) Construct the objective function E(R, t): Among them, p i represents a point in the template workpiece point cloud, q i represents a point in the measured matching point set, k i represents the normal vector of q i , n represents the number of points in the template workpiece point cloud; h j represents a point in the template edge point cloud, g j represents a point in the edge matching point set, m represents the number of points in the template edge point cloud; λ is a preset weight value; Use the optimization method to iteratively solve the objective function E(R, t) to obtain the rotation matrix R and the translation matrix t, and complete the registration between the point cloud of the template workpiece and the measured workpiece point cloud.

2. The point cloud registration method combining edge features according to claim 1, wherein: The method for searching for the edge points in the point cloud is: In the point cloud, use the normal-based boundary extraction method to obtain the edge points.

3. The point cloud registration method combining edge features according to claim 1, characterized in that: When acquiring the point cloud of the reference workpiece and the point cloud of the actual scene, a two-dimensional image is also acquired at the same position; there is a one-to-one correspondence between the pixel points in the two-dimensional image and the respective three-dimensional points in the point cloud; Denote the two-dimensional image corresponding to the point cloud of the reference workpiece as the reference two-dimensional image; denote the two-dimensional image corresponding to the point cloud in the actual scene as the measured two-dimensional image; The method for searching for the edge points in the point cloud of the template workpiece is: Perform edge extraction in the reference two-dimensional image to obtain the pixel coordinates of each edge point, search for the three-dimensional points in the point cloud of the template workpiece corresponding to each pixel coordinate, denote them as edge points, and store them into the template edge point cloud; The method for searching for the edge points in the measured workpiece point cloud is: Perform edge extraction in the measured two-dimensional image to obtain the pixel coordinates of each edge point, search for the three-dimensional points in the measured workpiece point cloud corresponding to each pixel coordinate, denote them as edge points, and store them into the measured edge point cloud.

4. The point cloud registration method combining edge features according to claim 1, characterized in that: When acquiring the point cloud of the reference workpiece and the point cloud of the actual scene, a two-dimensional image is also acquired at the same position; there is a one-to-one correspondence between the pixel points in the two-dimensional image and the respective three-dimensional points in the point cloud; Denote the two-dimensional image corresponding to the point cloud of the reference workpiece as the reference two-dimensional image, and the two-dimensional image corresponding to the point cloud in the actual scene as the measured two-dimensional image; Denote the point cloud of the template workpiece as point cloud I, and store the pixel points corresponding to each point in point cloud I in the reference two-dimensional image into point set A; Use the following steps to search for the edge points in point cloud I: ① Respectively take each pixel point in point set A as the center point, search for the pixel points in the n'×n' neighborhood around the center point, store the three-dimensional points in point cloud I corresponding to the found pixel points as neighborhood points, and denote the point in point cloud I corresponding to the center point as the reference point; ② Calculate the distances between each neighborhood point and the reference point, and count the total number B of points with distance values less than the threshold and the total number C of points with distance values greater than or equal to 2 - 4 times the threshold. Determine whether B and C are within a preset interval. If so, store the reference point as an edge point; otherwise, do not store it. In step 1), record the measured workpiece point cloud as the new point cloud I', and store the pixel points corresponding to each point in the new point cloud I' in the measured two-dimensional image into the new point set A'. Use the new point cloud I' and the new point set A' to perform steps ① and ② to search for edge points.

5. The point cloud registration method combining edge features according to claim 4, characterized in that: The threshold is d is the average point spacing of the point cloud; n' = 3, 5, 7, or 9; the preset interval is:

6. The point cloud registration method combining edge features according to claim 1, wherein: While obtaining the reference workpiece point cloud and the actual scene point cloud, also collect a two-dimensional image at the same position; there is a one-to-one correspondence between the pixel points in the two-dimensional image and the three-dimensional points in the point cloud. Record the two-dimensional image corresponding to the reference workpiece point cloud as the reference two-dimensional image. Record the two-dimensional image corresponding to the point cloud in the actual scene as the measured two-dimensional image. In step 1), perform the following preprocessing on the actual scene point cloud to screen out the measured workpiece point cloud from the actual scene point cloud: Find each two-dimensional pixel point corresponding to the template workpiece point cloud in the reference two-dimensional image, and record the image area formed by all the found pixel points as area one; manually set a selection box in the reference two-dimensional image, where the number of pixel points within the selection box is greater than area one and completely covers area one; save the image area within the selection box as a feature map, and select at least four points on the feature map as feature points. Centering on each feature point respectively, frame a local small area in the reference two-dimensional image, and perform template matching in the measured two-dimensional image using each local small area respectively. Record the matched area as area two, and record the center point of area two as the adaptation point. Solve the homography matrix using the feature points and the adaptation points. Then use the homography matrix to transform a single two-dimensional pixel point on the feature map into the measured two-dimensional image to obtain its pixel coordinates in the measured two-dimensional image, and store the three-dimensional point corresponding to this pixel coordinate in the actual scene point cloud into the measured workpiece point cloud; traverse each two-dimensional pixel point on the feature map, use the homography matrix to transform the two-dimensional pixel points on the feature map into the measured two-dimensional image to obtain the respective pixel coordinates of the two-dimensional pixel points in the measured two-dimensional image, and store the three-dimensional points corresponding to the pixel coordinates in the actual scene point cloud into the measured workpiece point cloud respectively.

7. The method for point cloud registration combining edge features according to claim 6, wherein: When using the optimization method to iteratively solve the objective function E(R, t), the initial values of the rotation matrix R and the translation matrix t are obtained through the SVD decomposition of the homography matrix.

8. The point cloud registration method combining edge features according to claim 1, wherein: The nearest neighbor search method in step 1) includes the kd-tree search method or the octree search method. The optimization method in step 2) is the least squares method, the Gauss-Newton method, or the Levenberg-Marquardt method.

9. The point cloud registration method incorporating edge features according to claim 1, characterized in that: When using the optimization method to iteratively solve the objective function E(R, t), the initial values of the rotation matrix R and the translation matrix t are obtained through coarse registration, or the initial values of the rotation matrix R and the translation matrix t are set as the identity matrix.

Images