Point cloud registration method guided by overlapping region similarity

Abstract

In order to solve the problem that the existing point cloud registration method is difficult to realize accurate registration of the measurement point cloud with the characteristics of extremely low overlap, lack of features, noise, and different densities, a point cloud registration method guided by the similarity of the overlapping region is provided.The overlapping region of the measurement point cloud of the adjacent view is extracted first, the scale size of the overlapping region is made consistent through scale comparison adjustment, the influence of the initial roughness on the inaccuracy of the extraction of the overlapping region caused by the inherent geometric error of the measurement system can be reduced, the problem of inconsistent shape of the overlapping region caused by noise and different densities can be overcome through bidirectional interaction, then the probability registration is carried out on the overlapping region with consistent scale size and shape, the transformation from the source point cloud to the target point cloud of the coarse registration is obtained, finally the obtained transformation is applied to the measurement point cloud of the adjacent view, and the robust registration of the measurement point cloud with the characteristics of extremely low overlap, lack of features, noise, and different densities is realized.

Description

Technical Field

[0001] This invention belongs to the field of image data processing technology. Background Technology

[0002] Optical measurement, with its advantages of flexibility, efficiency, feasibility, and dense measurement points, holds promise for solving the challenge of rapid and accurate measurement of complex industrial products. However, due to the limitations of sensor measurement range, a single sensor cannot obtain a complete point cloud of the target in a single measurement. Therefore, it is often necessary to integrate one or more sensors onto a mechanical device capable of multi-view measurement, such as a machine tool, robot, or multi-axis platform, to form a measurement system capable of obtaining a complete point cloud of the target. Although the measurement system is calibrated before measurement, due to the inherent geometric errors of the measurement system, the measurement point clouds acquired by the system from different viewpoints are difficult to perfectly align after being unified to a common coordinate system (coarse registration), resulting in unavoidable large stitching errors that affect the accuracy of optical measurement. Therefore, it is usually ensured that the measurement ranges of adjacent viewpoints have a certain overlap so that the measurement point clouds of adjacent viewpoints have a certain overlapping area. In this case, after unifying the measurement point clouds of each viewpoint to a common coordinate system, point cloud registration (fine registration) is performed on the measurement point clouds of adjacent viewpoints to reduce the stitching error.

[0003] Currently, mainstream point cloud registration methods fall into three categories: distance-based, probability-based, and feature-based. Among distance-based methods, the Iterative Nearest Point Registration (ICP) algorithm and its variants are widely used in various fields due to their simplicity and speed. However, their binary evaluation method makes them sensitive to initial pose, noise, and density variations in the point cloud, easily leading to local optima. Probability-based registration methods, represented by the Coherent Point Shift algorithm and the GMMReg algorithm, can effectively handle noisy and density-varying point clouds, but their registration accuracy needs further improvement. Feature-based methods extract point cloud features to guide the corresponding assignment, but these methods are not always effective when point cloud features are not significant. In recent years, driven by deep learning technology, a large number of point cloud registration networks have emerged, aiming to extract deep features to guide registration through powerful abstract information learning capabilities, or to enhance robustness by improving noise recognition capabilities. However, the generalization ability of these methods is affected by the quality of training data, and obtaining high-quality training data is difficult and costly.

[0004] Overall, the point cloud registration methods described above perform well in specific scenarios. However, in the field of aero-engines, in pursuit of higher service performance, aero-engine components are developing towards thinner walls and more complex technologies. These structural characteristics result in significantly lower overlap (overlap rate far below 50%) in their measured point clouds compared to those of other industrial products, along with features lacking, noise, and inconsistent density. For such measured point clouds, the aforementioned point cloud registration methods struggle to establish effective correspondences between the point clouds to be registered, thus failing to complete the task of precise point cloud registration.

[0005] Therefore, research on registration methods for measurement point clouds with characteristics such as extremely low overlap, lack of features, noise, and inconsistent density is of great significance and can provide important technical support for the accurate measurement of thin-walled complex components of aero-engines. Summary of the Invention

[0006] To address the technical problem that existing point cloud registration methods struggle to achieve accurate registration of measurement point clouds with characteristics such as extremely low overlap, lack of features, noise, and inconsistent density, this invention proposes a point cloud registration method guided by the similarity of overlapping regions.

[0007] The present invention adopts the following technical solution:

[0008] The point cloud registration method guided by the similarity of overlapping regions is characterized by the following steps:

[0009] Step 1: Define the first The first perspective and the first The measured point clouds obtained from each viewpoint are the target point cloud and the source point cloud. The source point cloud is coarsely registered to the target point cloud to obtain the coarsely registered source point cloud.

[0010] Step 2: Based on the endpoints of the target point cloud and the endpoints of the coarsely registered source point cloud, use the nearest neighbor search algorithm to extract the first source overlapping region from the coarsely registered source point cloud and the first target overlapping region from the target point cloud.

[0011] Step 3: Registration guided by similarity of overlapping regions;

[0012] Step 3.1: Based on scale comparison, adjust the scale of the first source overlapping region and the first target overlapping region to obtain a second source overlapping region and a second target overlapping region with the same scale.

[0013] Step 3.2: Through bidirectional interaction, adjust the shape of the second source overlap region and the second target overlap region to obtain a third source overlap region and a third target overlap region with the same scale and shape;

[0014] Step 3.3: Use a probability-based registration algorithm to finely register the third source overlapping region to the third target overlapping region, thus obtaining the transformation from the coarsely registered source point cloud to the target point cloud;

[0015] Step 4: Point cloud fine-tuning;

[0016] The transformation is applied to the coarsely registered source point cloud to finely register the coarsely registered source point cloud to the target point cloud, thus obtaining the finely registered source point cloud and completing the point cloud registration.

[0017] Furthermore, the nearest neighbor search algorithm in step 2 is the Kd-tree algorithm.

[0018] Furthermore, step 3.1 specifically includes:

[0019] Step 3.1.1: Solve for the directed bounding boxes of the first source overlapping region and the first target overlapping region respectively. Based on the directed bounding boxes, calculate the scale difference between the first source overlapping region and the first target overlapping region using the following formula. :

[0020] ;

[0021] In the formula, a, b, and c are the weights of the length, width, and height of the directed bounding box when calculating the scale difference, respectively. The values ​​of a, b, and c are all in the range of 0 to 1, and a + b + c = 1; d1, d2, and d3 are the length, width, and height of the directed bounding box of the first source overlapping region, respectively; d4, d5, and d6 are the length, width, and height of the directed bounding box of the first target overlapping region, respectively.

[0022] Step 3.1.2: Calculate the scale difference. Compared with the set scale difference threshold In comparison, if Delete the endpoint of the first source overlapping region; if Delete the starting point of the overlapping region of the first target;

[0023] Step 3.1.3: Repeat steps 3.1.1-3.1.2 until the scale difference is reached. satisfy At that time, a second source overlap region and a second target overlap region with the same scale are obtained;

[0024] The The value range is 0.001-0.01mm.

[0025] Furthermore, step 3.2 specifically includes:

[0026] Step 3.2.1: Construct the nearest point set;

[0027] Using the nearest neighbor search algorithm, the nearest points corresponding to each point in the second source overlapping region are searched in the second target overlapping region, and the nearest point set of the second source overlapping region is constructed.

[0028] Using the nearest neighbor search algorithm, the nearest points corresponding to each point in the second target overlapping region are searched in the second source overlapping region, and the nearest point set of the second target overlapping region is constructed.

[0029] Step 3.2.2: Solve for the source intermediate point cloud and the target intermediate point cloud;

[0030] Each point in the second source overlapping region corresponds one-to-one with each point in the nearest point set of the second source overlapping region, forming a point pair, which together constitute... Set of points, solve The midpoints of each pair of points in the set of points form the source intermediate point cloud; The number of points in the point cloud of the second source overlapping region; the number of points in the point cloud of the source middle region is equal to the number of points in the point cloud of the second source overlapping region;

[0031] Each point in the overlapping region of the second target corresponds one-to-one with each point in the nearest point set of the overlapping region of the second target, forming a point pair, which together constitute... Set of points, solve The midpoints of each pair of points in the set of points constitute the target intermediate point cloud; The number of points in the point cloud of the overlapping region of the second target; the number of points in the point cloud of the middle target is equal to the number of points in the point cloud of the overlapping region of the second target;

[0032] Step 3.2.3: Centroid transformation and point cloud fusion;

[0033] Centroid Transformation: Based on the difference between the centroid coordinates of the source intermediate point cloud and the centroid coordinates of the second source overlapping region, centroid transformation is performed on all points outside the start and end points within the source intermediate point cloud to eliminate the pose difference between the source intermediate point cloud and the second source overlapping region; Based on the difference between the centroid coordinates of the target intermediate point cloud and the centroid coordinates of the second target overlapping region, centroid transformation is performed on all points outside the start and end points within the target intermediate point cloud to eliminate the pose difference between the target intermediate point cloud and the second target overlapping region.

[0034] Point cloud fusion: The coordinates of the start and end points of the source intermediate point cloud are replaced with the coordinates of the start and end points of the second source overlapping region, respectively, thereby adding the pose information of the second source overlapping region to the source intermediate point cloud; The coordinates of the start and end points of the target intermediate point cloud are replaced with the coordinates of the start and end points of the second target overlapping region, respectively, thereby adding the pose information of the second target overlapping region to the target intermediate point cloud.

[0035] After centroid transformation and point cloud fusion, the source intermediate point cloud is transformed to the second source overlapping region, and the target intermediate point cloud is transformed to the second target overlapping region, resulting in a third source overlapping region and a third target overlapping region with the same scale and shape.

[0036] Furthermore, the nearest neighbor search algorithm described in step 3.2.1 is the Kd-tree algorithm.

[0037] Furthermore, in step 3.2.2:

[0038] The midpoints of all pairs of points are calculated using the following formula:

[0039] ;

[0040] ;

[0041] ;

[0042] in, , , for In the group of points, the first The coordinates of the midpoint of the pair of points. , , For the second source overlapping region The coordinates of the points , , For the nearest point set of the second source overlapping region, the first point of the second source overlapping region The coordinates of the points corresponding to each point;

[0043] The midpoints of all pairs of points are calculated using the following formula:

[0044] ;

[0045] ;

[0046] ;

[0047] in, , , for In the group of points, the first The coordinates of the midpoint of the pair of points. , , For the second target overlapping region The coordinates of the points , , For the nearest point set of the overlapping region of the second target, the first point of the overlapping region of the second target The coordinates of the points corresponding to each point.

[0048] Furthermore, the method for centroid transformation in step 3.2.3 is as follows:

[0049] Let the points outside the starting and ending points within the source intermediate point cloud be denoted as... n takes the values ​​2, 3, ... Point Perform the centroid transformation according to the following formula:

[0050] ;

[0051] ;

[0052] ;

[0053] in, , , Let be the coordinates of the centroid of the source midpoint cloud. , , The coordinates of the centroid of the second source overlapping region;

[0054] Let the points outside the starting and ending points within the target intermediate point cloud be denoted as... m takes values ​​of 2, 3, ... Point Perform the centroid transformation according to the following formula:

[0055] ;

[0056] ;

[0057] ;

[0058] in, , , Let the coordinates of the centroid of the target midpoint cloud be given. , , The coordinates of the centroid of the overlapping region of the second target.

[0059] Furthermore, the probability-based registration algorithm described in step 3.3 is a coherent point drift algorithm.

[0060] The beneficial effects of this invention are as follows:

[0061] 1. Due to the insignificant feature characteristics of the measurement point cloud of complex thin-walled aerospace components, this invention employs an overlap-based registration strategy. Specifically, it extracts the overlapping regions of adjacent viewpoint measurement point clouds after coarse registration, obtaining a first source overlapping region and a first target overlapping region. Similarity-guided registration is then performed on the point clouds of the first source overlapping region and the first target overlapping region to obtain a transformation from the coarsely registered source point cloud to the target point cloud. This transformation is then applied to the coarsely registered source point cloud, thereby enhancing the invention's ability to handle registration of point clouds with extremely low overlap. Secondly, this invention adjusts the scale of the first source overlapping region and the first target overlapping region through scale comparison, obtaining a second source overlapping region and a second target overlapping region with consistent scale. First, the invention reduces the impact of initial coarsening caused by inherent geometric errors in the measurement system on inaccurate extraction of overlapping regions. Second, to address the problem of inconsistent shapes in overlapping regions due to noise and density variations, the invention employs bidirectional interaction to adjust the shapes of the second source overlapping region and the second target overlapping region, resulting in a third source overlapping region and a third target overlapping region with consistent scale and shape. Finally, a probability-based registration algorithm guides the fine registration of the third source overlapping region and the third target overlapping region with consistent scale and shape, thereby obtaining a transformation from the coarsely registered source point cloud to the target point cloud. Ultimately, robust registration of measurement point clouds with extremely low overlap, lack of features, noise, and inconsistent density is achieved.

[0062] 2. Practical verification shows that the registration success rate and registration error of the present invention are superior to those of existing mainstream registration methods. Attached Figure Description

[0063] Figure 1 This is a flowchart of the method of the present invention.

[0064] Figure 2 This refers to the target point cloud and the source point cloud after coarse registration in this embodiment of the invention.

[0065] Figure 3 This refers to the first source overlapping region and the first target overlapping region in this embodiment of the invention.

[0066] Figure 4 This refers to the second source overlap region and the second target overlap region obtained after adjusting the scale size through scale comparison in this embodiment of the invention.

[0067] Figure 5 This refers to the third source overlap region and the third target overlap region obtained after bidirectional interactive shape adjustment in this embodiment of the invention.

[0068] Figure 6 These are the target point cloud and the finely registered source point cloud in this embodiment of the invention.

[0069] Explanation of reference numerals in the attached figures:

[0070] 1-Target point cloud; 2-Source point cloud after coarse registration; 3-Starting point of the source point cloud after coarse registration; 4-Ending point of the target point cloud; 5-First nearest point; 6-Second nearest point; 7-First source overlapping region; 8-First target overlapping region; 9-Second source overlapping region; 10-Second target overlapping region; 11-Third source overlapping region; 12-Third target overlapping region; 13-Source point cloud after fine registration. Detailed Implementation

[0071] The point cloud registration method of the present invention will be described in detail below with reference to the accompanying drawings, taking the measurement point cloud (which is a two-dimensional point cloud) obtained by a line laser multi-view measurement system for a thin-walled complex component in an aero-engine as an example.

[0072] For ease of description, the first point in the measurement point cloud of a thin-walled complex component acquired by a line laser multi-view measurement system is used. The measurement point cloud (i.e., the first measurement point cloud) The measured point cloud obtained from the first viewpoint is denoted as the source point cloud, and the first point cloud is denoted as the second point cloud. The measurement point cloud (i.e., the first measurement point cloud) The measured point cloud obtained from each viewpoint is denoted as the target point cloud. The indexes of each point in both the source point cloud and the target point cloud increase in a clockwise direction.

[0073] Reference Figure 1 The specific steps for registering the source point cloud to the target point cloud using the method of the present invention are as follows:

[0074] Step 1: Coarse registration of point clouds;

[0075] After the calibration of the line laser multi-view measurement system is completed, the transformation relationships between the coordinate systems within the system are known. Therefore, the source point cloud can be transformed based on these relationships to coarsely register it to the target point cloud 1, resulting in the coarsely registered source point cloud 2. (See Appendix) Figure 2 .

[0076] Step 2: Based on the endpoints of the source point cloud and the target point cloud after coarse registration, the overlapping region is extracted from both the source point cloud and the target point cloud using the nearest neighbor search algorithm.

[0077] In multi-view measurement, point cloud acquisition is sequential, and the endpoints and starting points of the target point cloud 1 and the coarsely registered source point cloud 2 are known.

[0078] Using a nearest neighbor search algorithm (such as the Kd-tree algorithm), the nearest point corresponding to the endpoint 4 of the target point cloud is searched in the coarsely registered source point cloud 2, and this point is recorded as the second nearest point 6. The first source overlapping region 7 can be obtained by extracting the point cloud between the starting point 3 of the coarsely registered source point cloud and the second nearest point 6 (see appendix). Figure 3), denoted as N represents the first source overlapping region. The number of points in the point cloud.

[0079] Using a nearest neighbor search algorithm (such as the Kd-tree algorithm), the nearest point corresponding to the starting point 3 of the source point cloud after coarse registration is searched in the target point cloud 1, and this point is recorded as the first nearest point 5. The first target overlapping region 8 can be obtained by extracting the point cloud between the first nearest point 5 and the ending point 4 of the target point cloud (see appendix). Figure 3 ), denoted as M represents the overlapping area of ​​the first target. The number of points in the point cloud;

[0080] Step 3: Registration guided by similarity of overlapping regions;

[0081] Step 3.1: Based on scale comparison, adjust the scale of the first source overlapping region and the first target overlapping region to obtain a second source overlapping region and a second target overlapping region with the same scale.

[0082] Step 3.1.1 Calculate the scale difference between the first source overlap region and the first target overlap region;

[0083] The first source overlap region was solved using principal component analysis. Directed bounding box Overlapping area with the first target Directed bounding box , towards the enclosure The length, width, and height of the directed bounding box are denoted as d1, d2, and d3, respectively. The length, width, and height are denoted as d4, d5, and d6, respectively. The first source overlapping region is calculated using the following formula. Overlapping area with the first target scale difference :

[0084] ;

[0085] In the formula, a, b, and c represent the weights of the length, width, and height of the directed bounding box when calculating the scale difference, respectively. The values ​​of a, b, and c are all in the range of 0 to 1, and a + b + c = 1. In this embodiment, a = 0.5, b = 0.5, and c = 0. Furthermore, since the measurement point cloud acquired by the line laser multi-view measurement system is a two-dimensional point cloud, the directed bounding box here is a two-dimensional bounding box; therefore, d3 and d6 are both 0.

[0086] Step 3.1.2 Adjust the scale of the first source overlapping region and / or the first target overlapping region;

[0087] The scale difference calculated in step 3.1.1 Compared with the set scale difference threshold In comparison, among which, The value range is 0.001-0.01mm.

[0088] like Delete the first source overlapping region The endpoint, after deletion, the first source overlapping region Decrease the current number of points in the point cloud by 1;

[0089] like Delete the overlapping area of ​​the first target The starting point, after deletion, the first target overlapping area Decrease the current number of points in the point cloud by 1.

[0090] Step 3.1.3 Repeat steps 3.1.1-3.1.2 until the scale difference is calculated. satisfy At that time, a second source overlapping region 9 and a second target overlapping region 10 with the same scale were obtained, as shown in the figure. Figure 4 As shown.

[0091] The second source overlapping region 9 is denoted as :

[0092] ;

[0093] in, Second source overlapping region The Middle One point, Second source overlapping region Point cloud point count, , For the first source overlapping region The number of times the scale adjustment is performed;

[0094] The overlapping area of ​​the second target is denoted as 10. :

[0095] ;

[0096] in, For the second target overlapping area The Middle One point, For the second target overlapping area Point cloud point count, , For the overlapping area of ​​the first target The number of times the scale adjustment is performed.

[0097] Step 3.2: Adjust the shape of the second source overlap region and the second target overlap region through bidirectional interaction to obtain a third source overlap region and a third target overlap region with the same scale and shape;

[0098] Step 3.2.1: Construct the nearest point set;

[0099] Using a nearest neighbor search algorithm (such as the Kd-tree algorithm) in the overlapping region of the second target In the middle, find the second source overlapping region. The nearest points corresponding to each point within the region are defined as the set of nearest points of the second source overlapping region. :

[0100] ;

[0101] In the second source overlap region, a nearest neighbor search algorithm (such as the Kd-tree algorithm) is used. In the middle, find the overlapping area of ​​the second target. The nearest points corresponding to each point within the area are defined as the set of nearest points in the overlapping region of the second target. :

[0102] .

[0103] Step 3.2.2: Solve for the source intermediate point cloud and the target intermediate point cloud;

[0104] The second source overlap region The first Each point is represented as The set of nearest points in the second source overlapping region The first Each point is represented as Due to the second source overlap region Its nearest set The points have the same number of points and each point corresponds to a point pair, forming a total of 10 points. Set of points, solve The midpoints of each pair of points in the set of points form the source intermediate point cloud, denoted as . ;

[0105] in, Source intermediate point cloud The first in One point, ; , ;

[0106] The overlapping area of ​​the second target The first Each point is represented as The set of nearest points in the overlapping region of the second target The first Each point is represented as Due to the overlapping area of ​​the second target Its nearest set The points have the same number of points and each point corresponds to a point pair, forming a total of 10 points. Set of points, solve The midpoints of each pair of points in the set of points form the target intermediate point cloud, denoted as . ;

[0107] in, For the target intermediate point cloud The first in One point, ; , .

[0108] Step 3.2.3: By transforming the centroid and fusing the point cloud, the source intermediate point cloud is transformed into the second source overlapping region, and the target intermediate point cloud is transformed into the second target overlapping region, so as to obtain a third source overlapping region and a third target overlapping region with the same scale and shape.

[0109] Step 3.2.3.1 Convert the source intermediate point cloud Transform to the second source overlap region ;

[0110] For source intermediate point cloud Its pose overlaps with the region of the second source. Since there are differences, we analyze the source intermediate point cloud based on the difference in centroid coordinates between the two. To eliminate pose differences between points outside the inner starting and ending points, a centroid transformation is performed; point cloud fusion is then performed, specifically the source and intermediate point clouds. The starting and ending points are respectively set as the second source overlapping region. The starting and ending points, with the source midpoint cloud Add a second source overlap region The pose information.

[0111] The second source overlap region Intermediate point cloud of source The centroids are denoted as:

[0112] ;

[0113] ;

[0114] Source intermediate point cloud Points within The specific steps for centroid transformation and point cloud fusion are as follows:

[0115] When n=1, point , as the source intermediate point cloud The starting point in the cloud will be the source intermediate point. The coordinates of the starting point are set as the second source overlapping region. The coordinates of the starting point, i.e.:

[0116] ;

[0117] ;

[0118] When 1 <n< When n takes the values ​​2, 3, ..., At time, point Source intermediate point cloud All points other than the starting and ending points are determined by the centroid. With the center of mass For the source intermediate point cloud Points other than the starting and ending points are processed as follows:

[0119] ;

[0120] ;

[0121] when At time, point , as the source intermediate point cloud The endpoint in the cloud will be the source intermediate point. The coordinates of the endpoint in the middle are set as the second source overlapping region. The coordinates of the endpoint, i.e.:

[0122] ;

[0123] .

[0124] After the above centroid transformation and point cloud fusion processing, the source intermediate point cloud can be... Transform to the second source overlap region The third source overlap region 11 is obtained, as shown in the attached figure. Figure 5 As shown.

[0125] Step 3.2.3.2: Create a cloud of the target intermediate points. Transform to the second target overlapping region ;

[0126] For the target intermediate point cloud Its pose overlaps with the area of ​​the second target. Since there are differences, we use the difference in centroid coordinates between the two to determine the target intermediate point cloud. To eliminate pose differences between points outside the inner starting and ending points, a centroid transformation is performed; point cloud fusion is then performed, specifically the point cloud of the target's middle layer. The starting and ending points are respectively set as the overlapping areas of the second target. The starting point and the ending point, to the midpoint of the target cloud. Add a second target overlapping area The pose information.

[0127] The overlapping area of ​​the second target Midpoint cloud of the target The centroids are denoted as:

[0128] ;

[0129] ;

[0130] For the target intermediate point cloud Points within The specific steps for centroid transformation and point cloud fusion are as follows:

[0131] When m=1, point , for the target intermediate point cloud The starting point in the cloud is the target midpoint. The coordinates of the starting point are set as the overlapping area of ​​the second target. The coordinates of the starting point, i.e.:

[0132] ;

[0133] ;

[0134] When 1 <m< That is, m takes the values ​​2, 3, ... At time, point For the target intermediate point cloud All points other than the starting and ending points are determined by the centroid. With the center of mass For the target intermediate point cloud Points other than the starting and ending points are processed as follows:

[0135] ;

[0136] ;

[0137] When m= At time, point , for the target intermediate point cloud The endpoint in the cloud will be the midpoint of the target. The coordinates of the endpoint are set as the overlapping area of ​​the second target. The coordinates of the endpoint, i.e.:

[0138] ;

[0139] .

[0140] After the above centroid transformation and point cloud fusion processing, the target's central point cloud can be obtained. Transform to the second target overlapping region The third target overlapping region 12 was obtained, as shown in the attached figure. Figure 5 As shown.

[0141] The aforementioned third source overlapping region 11 and the third target overlapping region 12 are identical in size and shape.

[0142] The order of steps 3.2.3.1 and 3.2.3.2 above can be interchanged.

[0143] Step 3.3: Obtain the transformation from the coarsely registered source point cloud to the target point cloud through probabilistic registration;

[0144] By using a probabilistic registration algorithm (a known method, such as the coherent point drift algorithm), the third source overlapping region 11 is finely registered to the third target overlapping region 12, and the transformation from the coarsely registered source point cloud 2 to the target point cloud 1 can be obtained. This transformation is a homogeneous matrix, which contains a rotation matrix and a translation vector.

[0145] Step 4: Point cloud fine-tuning;

[0146] The transformation obtained in step 3.3 is applied to the coarsely registered source point cloud 2, so that the coarsely registered source point cloud 2 is finely registered to the target point cloud 1, resulting in the finely registered source point cloud 13. At this point, the point cloud registration is complete, as shown in the attached figure. Figure 6 As shown.

[0147] The above description of the invention is based on a two-dimensional point cloud. The registration method for three-dimensional point clouds is the same as that for two-dimensional point clouds, except that the z-coordinate value of the points involved is not equal to 0.

[0148] Technical effectiveness verification:

[0149] The registration performance of this invention and four existing mainstream registration methods was compared in 28 test datasets. The success rate of these five methods in completing the registration task and the proportion of achieving the lowest registration error in the 28 test datasets are statistically analyzed, as shown in the table below.

[0150]

[0151] The results show that the success rate of the present invention in completing the registration task is 28 / 28, and the proportion of achieving the lowest registration error is 25 / 28, which is significantly better than the other four mainstream registration methods.

Claims

1. A method of point cloud registration guided by overlapping region similarity, characterized in that, The method comprises the following steps: Step 1: defining the measurement point clouds obtained in the ith view and the i+1th view as a target point cloud and a source point cloud respectively, coarsely registering the source point cloud to the target point cloud to obtain a coarsely registered source point cloud; Step 2: based on the endpoints of the target point cloud and the endpoints of the coarsely registered source point cloud, using a nearest neighbor search algorithm, extracting a first source overlap region from the coarsely registered source point cloud and a first target overlap region from the target point cloud; Step 3: overlap region similarity guided registration; Step 3.1: based on scale comparison, adjusting the scale sizes of the first source overlap region and the first target overlap region to obtain second source overlap region and second target overlap region with consistent scale sizes; Step 3.1.1: solving the oriented bounding boxes for the first source overlap region and the first target overlap region, respectively, calculating the scale difference for the first source overlap region and the first target overlap region based on the oriented bounding boxes using the following equation : ； In the formula, a, b, c are weights of the length, width and height of the oriented bounding box when calculating the scale difference, the value ranges of a, b and c are all 0-1, and a+b+c=1; d1, d2 and d3 are the length, width and height of the oriented bounding box of the first source overlap region; d4, d5 and d6 are the length, width and height of the oriented bounding box of the first target overlap region; Step 3.1.2: comparing the scale difference with a set scale difference threshold If , the end point of the first source overlap region is deleted; If , delete the start point of the first target overlapping region; Step 3.1.3: Repeat steps 3.1.1-3.1.2 until the scale difference is satisfied satisfies , the second source overlap region and the second target overlap region with consistent scale size are obtained; The ranges from 0.001 to 0.01 mm; Step 3.2: by bidirectional interaction, adjusting the shapes of the second source overlap region and the second target overlap region to obtain third source overlap region and third target overlap region with consistent scale sizes and shapes; Step 3.3: by a probability-based registration algorithm, precisely registering the third source overlap region to the third target overlap region to obtain a transformation from the coarsely registered source point cloud to the target point cloud; Step 4: point cloud precise registration; Applying the transformation to the coarsely registered source point cloud, precisely registering the coarsely registered source point cloud to the target point cloud to obtain a precisely registered source point cloud, and completing the point cloud registration.

2. The method of claim 1, wherein, The nearest neighbor search algorithm in step 2 is a Kd-tree algorithm.

3. The method of claim 2, wherein, Step 3.2 is specifically: Step 3.2.1: constructing a nearest point set; By using a nearest neighbor search algorithm, searching for the nearest points corresponding to each point in the second source overlap region in the second target overlap region to construct a nearest point set of the second source overlap region; By using a nearest neighbor search algorithm, searching for the nearest points corresponding to each point in the second target overlap region in the second source overlap region to construct a nearest point set of the second target overlap region; Step 3.2.2: solving a source intermediate point cloud and a target intermediate point cloud; The point pairs are formed by one-to-one correspondence between each point in the second source overlapping area and each point in the nearest point set of the second source overlapping area, and a total of solving the group of point pairs the midpoints of each group of point pairs in the group of point pairs, and the midpoints form a source midpoint cloud; the number of points in the point cloud of the second source overlapping area; the number of points in the source midpoint cloud is equal to the number of points in the point cloud of the second source overlapping area; The point pairs are formed by one-to-one correspondence between each point in the second target overlapping area and each point in the nearest point set of the second target overlapping area, and a total of solving the group of point pairs the midpoints of each group of point pairs in the group of point pairs, and the midpoints form the target intermediate point cloud; the number of points in the point cloud of the second target overlapping area; the number of points in the target intermediate point cloud is equal to the number of points in the point cloud of the second target overlapping area; Step 3.2.3: centroid transformation and point cloud fusion; Centroid transformation: based on the difference between the centroid coordinates of the source intermediate point cloud and the centroid coordinates of the second source overlap region, performing centroid transformation on each point in the source intermediate point cloud except the start point and the end point to eliminate the pose difference between the source intermediate point cloud and the second source overlap region; Based on the difference between the centroid coordinates of the target intermediate point cloud and the centroid coordinates of the second target overlap region, performing centroid transformation on each point in the target intermediate point cloud except the start point and the end point to eliminate the pose difference between the target intermediate point cloud and the second target overlap region; Point cloud fusion: replacing the coordinates of the start point and the end point of the source intermediate point cloud with the coordinates of the start point and the end point of the second source overlap region, thereby adding the pose information of the second source overlap region in the source intermediate point cloud; The coordinates of the start point and the end point of the target intermediate point cloud are replaced by the coordinates of the start point and the end point of the second target overlap region respectively, so as to add the pose information of the second target overlap region in the target intermediate point cloud; After the centroid transformation and the point cloud fusion, the source intermediate point cloud is transformed to the second source overlap region, and the target intermediate point cloud is transformed to the second target overlap region, so as to obtain the third source overlap region and the third target overlap region which are consistent in size and shape.

4. The method of claim 3, wherein, The nearest neighbor search algorithm described in step 3.2.1 is a Kd-tree algorithm.

5. The overlap region similarity guided point cloud registration method according to claim 3 or 4, characterized in that, In step 3.2.2: The midpoints of the group point pairs are each calculated as follows: ； ； ； wherein, , , is coordinates of the midpoint of the coordinates of the midpoint of the , , coordinates of the coordinates of the , , coordinates of the point corresponding to the coordinates of the point corresponding to the The midpoints of the group point pairs are each calculated as follows: ； ； ； wherein , , is the coordinate of the midpoint of the coordinate of the midpoint of the , , is the coordinate of the point in the second target overlapping region, , , is the coordinate of the point corresponding to the point in the second target overlapping region.

6. The method of claim 5, wherein, The method of centroid transformation in step 3.2.3 is specifically: Points in the source intermediate point cloud outside the start point and the end point are denoted as , n takes 2, 3,..., respectively, , the point is centroid-transformed according to the following formula: ； ； ； wherein, , , is the coordinate of the centroid of the source intermediate point cloud, , , is the coordinate of the centroid of the second source overlap region; Points in the target intermediate point cloud outside the start point and the end point are denoted as , m is respectively 2, 3, …, , the point is centroid transformed according to the following formula: ； ； ； wherein, , , is a coordinate of a centroid of the target intermediate point cloud, , , is a coordinate of a centroid of the second target overlap region.

7. The method of claim 6, wherein, The probability-based registration algorithm described in step 3.3 is a coherent point drift algorithm.

Images