A sheep body three-dimensional point cloud registration method and system based on a local overlapping area

Abstract

The application discloses a sheep three-dimensional point cloud registration method and system based on a local overlapping area. The application estimates the proportion relationship between the overlapping area and the whole point cloud of the target detection body covered by the vertical camera and the horizontal camera by uniformly down-sampling the filtered point cloud data, introduces the judgment of the chamfer distance in the CPD algorithm, obtains the transformation matrix of the local overlapping area, applies the local transformation matrix to the corresponding rotation matrix and translation vector changes of the left and right complete point clouds, and finally, the changed left view, right view and top view point clouds are merged to obtain the registered three-dimensional sheep point cloud. The application effectively solves the problem that the initialization position is sensitive and the asymmetric geometric shape is trapped in a local optimal solution, thereby causing the registration failure. Moreover, the registration speed is improved by 3 times compared with the CPD algorithm before the improvement because the chamfer distance is introduced.

Description

Technical Field

[0001] This invention belongs to the technical field of three-dimensional point cloud registration, specifically relating to a method and system for three-dimensional point cloud registration of sheep bodies based on locally overlapping regions. Background Technology

[0002] 3D point cloud registration technology refers to the process of transforming multiple point cloud data containing the same object from different spatial coordinate systems to the same coordinate system using a certain transformation matrix. Since a single depth camera usually cannot capture the entire target when its position is fixed, it is necessary to stitch together the point clouds of overlapping areas from multiple depth cameras into a panoramic point cloud data with a larger field of view for further scientific experiments.

[0003] Existing point cloud registration algorithms such as ICP (Iterative Closest Point), RPM (Robust Point Matching), KC (K-4PCS), and CPD (Coherent point drift) are sensitive to the initial position. They require manual coarse registration of the source and target point clouds before fine registration, and are prone to getting trapped in local optima, resulting in incomplete point cloud registration. Summary of the Invention

[0004] To address the issues of low accuracy and slow registration speed in existing point cloud registration methods, this invention provides a method and system for registering sheep 3D point clouds based on locally overlapping regions. By using the registration method based on overlapping regions of point clouds, the low accuracy of point clouds without significant features during overall registration is avoided. Furthermore, due to the local registration method, it has a faster registration speed compared to global registration.

[0005] According to one aspect of the present invention, a method for registering three-dimensional point clouds of sheep bodies based on locally overlapping regions is provided, comprising:

[0006] Obtain the point cloud data of the sheep body from the top view, left view, and right view respectively;

[0007] The point cloud data of the sheep body in the top view, left view and right view are segmented locally to obtain the overlapping area of ​​the point cloud;

[0008] Using the point cloud data of the left and right views as the source point cloud and the point cloud data of the top view as the target point cloud, an improved CPD algorithm is introduced to register the target point cloud and the source point cloud in the overlapping area of ​​the point cloud, thereby obtaining the local point cloud rigidity changes of the left and right views; wherein, the improved CPD algorithm introduces the chamfer distance judgment.

[0009] Based on the rigidity changes of the local point clouds in the left and right views, the complete point clouds in the left and right views are transformed by rotation matrix and translation vector to obtain the rigidity changes of the complete point clouds in the left and right views.

[0010] The top view is merged with the transformed left and right views to obtain the registered 3D sheep point cloud.

[0011] As a further technical solution, after obtaining the sheep body point cloud data from the top, left, and right views, the solution also includes:

[0012] The point cloud data of the sheep body in the top view, left view, and right view were downsampled respectively.

[0013] As a further technical solution, after downsampling the sheep body point cloud data from the top, left, and right views, the solution also includes:

[0014] Utilize the 3D perspective of multiple cameras to estimate possible overlapping areas of point clouds;

[0015] Based on the budget results, local point cloud segmentation is performed to obtain the overlapping area of ​​the point cloud.

[0016] As a further technical solution, an improved CPD algorithm is introduced to register the target point cloud and the source point cloud in the overlapping area of ​​the point cloud, including:

[0017] The source point cloud in the left or right view is subjected to iterative rigid changes, and convergence is judged when the number of iterations has not reached the specified value.

[0018] When the algorithm convergence between the target point cloud and the source point cloud fails to reach the rated value, the chamfer distance is judged, and the rigidity change is adjusted for the next iteration.

[0019] If the chamfer distance between the target point cloud and the source point cloud does not reach the specified value, proceed to the next iteration;

[0020] When the number of iterations reaches the specified value, the algorithm converges to the specified value, or the chamfer distance reaches the specified value, the iteration exits and the local rigidity change of the left or right view is output.

[0021] As a further technical solution, the chamfer distance is determined, including:

[0022] Normalize the target point cloud and the source point cloud after rigid changes;

[0023] Based on the normalization results, determine whether the chamfer distance between the target point cloud and the source point cloud reaches the specified value. The chamfer distance is calculated as follows:

[0024]

[0025] Where X and Y' represent the target point cloud and the source point cloud after rigid transformation, respectively, and ||.||2 represents the Euclidean distance.

[0026] According to one aspect of the present invention, a three-dimensional point cloud registration system for sheep bodies based on locally overlapping regions is provided, comprising:

[0027] The first main module is used to acquire point cloud data of the sheep body from the top view, left view, and right view, respectively.

[0028] The second main module is used to perform local region segmentation on the acquired top view, left view and right view sheep body point cloud data respectively to obtain the point cloud overlapping area;

[0029] The third main module is used to take the point cloud data of the left and right views as the source point cloud and the point cloud data of the top view as the target point cloud. An improved CPD algorithm is introduced to register the target point cloud and the source point cloud in the overlapping area of ​​the point cloud, so as to obtain the local point cloud rigidity changes of the left and right views. The improved CPD algorithm introduces the chamfer distance judgment.

[0030] The fourth main module is used to perform rotation matrix and translation vector transformations on the complete point clouds of the left and right views based on the local point cloud rigidity changes of the left and right views, so as to obtain the complete point cloud rigidity changes of the left and right views.

[0031] The fifth main module is used to merge the top view with the transformed left and right views to obtain the registered 3D sheep point cloud.

[0032] According to one aspect of the present invention, a method and system for registering three-dimensional point clouds of sheep bodies based on locally overlapping regions are provided, comprising a memory and a processor, wherein the memory stores program instructions that are executed by the processor, and the processor invokes the program instructions to perform the steps of the method for registering three-dimensional point clouds of sheep bodies based on locally overlapping regions.

[0033] According to one aspect of the present invention, a non-transitory computer-readable storage medium is provided, the non-transitory computer-readable storage medium storing computer instructions that cause the computer to perform the steps of the described method for registering sheep three-dimensional point clouds based on locally overlapping regions.

[0034] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0035] This invention discloses a method and system for registering 3D point clouds of sheep based on locally overlapping regions. The method includes: uniformly downsampling the filtered point cloud data; estimating the proportional relationship between the overlapping regions of the target detection bodies covered by the vertical and horizontal cameras and the overall point cloud using the camera field of view; introducing the chamfer distance judgment into the CPD algorithm to obtain the transformation matrix of the locally overlapping regions; applying the local transformation matrix to perform corresponding rotation matrix and translation vector transformations on the left and right complete point clouds respectively; and finally merging the top view and the transformed left and right view point clouds respectively to obtain the registered 3D sheep point cloud.

[0036] This invention addresses the shortcomings of traditional algorithms in 3D point cloud registration, such as the need for coarse registration and sensitivity to initialization position. It effectively solves the problem of registration failure caused by initialization position sensitivity and getting trapped in local optima when dealing with asymmetric geometry. Furthermore, by introducing chamfer distance, the registration speed is increased by 3 times compared to the previous CPD algorithm. Attached Figure Description

[0037] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the accompanying drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the accompanying drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0038] Figure 1 This is a schematic diagram of the direct registration result when the traditional algorithm does not have coarse registration.

[0039] Figure 2 This is a schematic diagram illustrating the estimated proportion of overlapping areas provided in an embodiment of the present invention.

[0040] Figure 3 This is a schematic diagram of obtaining the overlapping region provided in an embodiment of the present invention.

[0041] Figure 4 This is a schematic diagram of global registration after local overlapping region registration provided in an embodiment of the present invention.

[0042] Figure 5 This is a schematic diagram of the sheep body 3D point cloud registration process based on locally overlapping regions, provided for an embodiment of the present invention. Detailed Implementation

[0043] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention. In addition, the technical features of the various embodiments or individual embodiments provided by the present invention can be arbitrarily combined to form new technical solutions. Such combinations are not bound by the order of steps and / or structural composition patterns, but must be based on the ability of those skilled in the art to implement them. When the combination of technical solutions is contradictory or cannot be implemented, it should be considered that such a combination of technical solutions does not exist and is not within the scope of protection claimed by the present invention.

[0044] Existing registration algorithms are sensitive to initialization. If the initial pose is not chosen properly, the algorithm is prone to getting trapped in local optima. In the presence of noise and outliers, it may lead to incorrect matching of corresponding points, thus affecting the registration accuracy. Figure 1 (As shown). Therefore, this invention improves the efficiency and generalization ability of registration by registering the locally overlapping areas when point clouds from different perspectives partially overlap, and then applying the transformation changes globally.

[0045] This invention provides a method for registering 3D point clouds of sheep bodies based on locally overlapping regions. This method segments the target detection body based on local region overlap, registers the local regions after segmentation, obtains the local transformation matrix, and then performs complete point cloud registration. This avoids the disadvantage of low accuracy when performing overall registration on point clouds without significant features. Furthermore, because it is a local registration, it has a faster registration speed compared to global registration.

[0046] like Figure 5 As shown, this invention provides a method for registering three-dimensional point clouds of sheep bodies based on locally overlapping regions, including the following steps:

[0047] Step 1: Acquire the sheep's body point cloud data from the top, left, and right views, respectively, and perform preprocessing. Preprocessing includes filtering and point cloud downsampling, that is, reducing the target point cloud and the original point cloud to a predetermined number of points through downsampling.

[0048] The advantage of this step is that by downsampling to a predetermined number of points, important features of the point cloud are preserved while improving the computational efficiency of subsequent point cloud registration.

[0049] Step 2: Perform local region segmentation on the sheep body point cloud data of the top view, left view, and right view to obtain the overlapping area of ​​the point cloud.

[0050] Preferably, step 2, which involves segmenting the target and source point cloud local regions, specifically includes the following sub-steps:

[0051] Step 2.1: Estimate the overlapping area of ​​the point cloud using the camera's field of view. Based on the camera parameters, obtain the vertical viewing angle α of the top camera and the horizontal viewing angle β of the other camera. The distance from the camera to the ground...

[0052] h is the distance from the camera to the outermost frame of the detection device, d is the angle between the top camera and the horizontal camera, and γ represents the ratio of the overlapping area to the sum of the areas covered by the top camera and the horizontal camera. The value of γ is estimated.

[0053] Top camera: The horizontal viewing angle is α. Looking down from the top, the field of view can be approximated as the base of a cone with a radius r. h It can be done Calculate the area covered by the top camera.

[0054] Horizontal camera: The vertical viewing angle is β. When viewed from left to right, the field of view can be approximated as the base of a cone with a radius r. v It can be done Calculate the area covered by the horizontal camera.

[0055] Maximum inscribed ellipse: The common coverage area can be an ellipse, where the major axis a of the ellipse approximates Max(r). h ,r v The minor axis b is approximately equal to Min(r). h ,r v The area of ​​the ellipse is A. o =π.ab,

[0056]

[0057] Step 2.2: Based on the budget results, the overlapping areas of the point cloud are segmented using point cloud segmentation technology to divide the overlapping areas by γ ratio, and the horizontal and top overlapping areas are segmented for subsequent registration.

[0058] Preferably, the point cloud of the left view is segmented along the Z-axis to obtain data of the overlapping area of ​​the left sheep body point cloud near the top. The point cloud of the right view is segmented along the Z-axis to obtain data of the overlapping area of ​​the right sheep body point cloud near the top.

[0059] The top view local region segmentation includes: 1) segmenting the point cloud of the top view along the X-axis to obtain point cloud data of the overlapping area of ​​the sheep body point cloud near the left; 2) segmenting the point cloud of the top view along the X-axis to obtain point cloud data of the overlapping area of ​​the sheep body point cloud near the right.

[0060] The advantage of this step is that registration based on overlapping point cloud regions avoids the problem of getting stuck in local optima during global registration, and because it is local registration, it has a faster registration speed compared to global registration.

[0061] Step 3: Using the point cloud data of the left and right views as the source point cloud and the point cloud data of the top view as the target point cloud, an improved CPD algorithm is introduced to register the target point cloud and the source point cloud in the overlapping area of ​​the point cloud, so as to obtain the local point cloud rigidity changes of the left and right views.

[0062] Step 3.1, Rigid Transformation of the Source Point Cloud: Apply the transformation matrix {R} from the previous iteration (Step 3.3.6) to the source point cloud Y obtained in Step 2. i-1 , t i-1 The transformation yields point cloud Y', where...

[0063] Step 3.2, Iteration count determination: Determine whether the number of iterations has reached the specified value. If it has, exit the iteration and proceed to step 4.

[0064] Step 3.3: Determine whether the algorithm convergence between the target point cloud in the local region and the source point cloud in the local region has reached the specified value. If it has, exit the iteration and proceed to step 4.

[0065] Furthermore, step 3.3 includes the following sub-steps:

[0066] Step 3.3.1, calculate the distance between point clouds: calculate the squared distance of all point pairs between the target point cloud X and the transformed source point cloud Y'. The result is an M x N distance matrix Q, where M is the number of points in Y' and N is the number of points in X.

[0067] Step 3.3.2, Calculate the constant term: Calculate the constant term c of the Gaussian distribution, which will be used for subsequent probability calculations.

[0068] Step 3.3.3, Calculate the probability matrix: Use the exponential term of the Gaussian distribution to transform the distance matrix into a probability matrix. Where σ 2 Q is the variance of the Gaussian distribution. ij Q is the distance matrix, representing the sum of squares of the distances between the i-th source point and the j-th target point.

[0069] Step 3.3.4, Normalizing the probability matrix: First, calculate the sum of each column of matrix P to obtain a vector d. Where P i,: Let d represent the i-th row of matrix P, and then copy vector d M times to form an M*N matrix: Next, replace all zero elements in matrix d with a very small positive number ∈: Finally, the constant c is added to each element of matrix d: d ij =d ij +c, use the denominator d to normalize the probability matrix P, thereby obtaining the normalized probability matrix.

[0070] Step 3.3.5, Calculate auxiliary variables:

[0071] i. Calculate the sum of each column of the probability matrix to obtain the total probability of each target point.

[0072] ii. Calculate the sum of each row of the probability matrix to obtain the total probability of each source point.

[0073] iii. Calculate the sum of the probability matrices, i.e., the total probability of all pairs of points.

[0074] Step 3.3.6, Adjust the rigidity transformation matrix: Based on the probability matrix P, calculate the mean of the target point cloud X and the source point cloud Y', center the point cloud, calculate the covariance matrix, perform singular value decomposition (SVD), and calculate the rotation and translation parameters to obtain the adjusted rigidity transformation matrix.

[0075] Step 3.3.7, obtain the convergence value of the algorithm: Apply the auxiliary variables obtained in step 3.3.5 and the change matrix obtained in step 3.3.6 to update the covariance of the Gaussian mixture model, and obtain the absolute difference value of the covariance before and after the update, which is used to determine the convergence value of the algorithm.

[0076] Step 3.4: Determine whether the chamfer distance between the target point cloud and the source point cloud in the local region reaches the specified value. If it does, exit the iteration and proceed to step 4.

[0077] Step 3.4 includes the following sub-steps:

[0078] Step 3.4.1, Point Cloud Normalization: Normalize the local source Y' and target point cloud X obtained in Step 3.1 for the calculation of chamfer distance.

[0079] Step 3.4.2, Point Cloud Chamfer Distance Calculation: Determine whether the chamfer distance between the target point cloud in the local area and the source point cloud in the local area reaches the specified value. The chamfer distance is calculated as follows:

[0080]

[0081] Where X and Y' represent the target point cloud and the source point cloud after rigid transformation, respectively, and ||.||2 represents the Euclidean distance.

[0082] The advantage of this step is that by adding the chamfer distance judgment, the registration effect of the algorithm can be significantly improved when dealing with complex scenes and asymmetrical geometry.

[0083] Step 3.5: Return the transformation matrix obtained in Step 3.3.6 to Step 3.1 for a new round of iteration.

[0084] Step 4: Obtain the local rigidity transformation matrix and apply rigidity transformation to the complete point cloud: Apply the local transformation matrix to perform corresponding rotation matrix and translation vector transformations on the left and right complete point clouds respectively.

[0085] The advantage of this step is that local registration is faster than global point cloud registration, while avoiding the local optimum problem caused by poor initial point cloud position, thus enhancing the robustness and adaptability of the algorithm.

[0086] Step 5, Merge left, right, and top views: Merge the point clouds of the top view and the transformed left and right views to obtain the registered 3D sheep body point cloud.

[0087] The registration algorithm of this invention, after adding the chamfer distance judgment, reduces the number of iterations from 70 to 25 to achieve the registration effect, and the registration time is reduced from 6.9137s to 2.5156s. At the same time, the registration method based on local overlapping areas will not cause the problem of global registration failure.

[0088] The implementation of the various embodiments of the present invention is based on programmed processing by a device with processor functionality. Therefore, in practical engineering, the technical solutions and functions of the various embodiments of the present invention are encapsulated into various modules. Based on this reality, and building upon the above embodiments, the embodiments of the present invention provide a sheep body 3D point cloud registration system based on locally overlapping regions. This system is used to execute a sheep body 3D point cloud registration method based on locally overlapping regions as described in the above method embodiments.

[0089] The system includes: a first main module for acquiring sheep body point cloud data from top, left, and right views respectively; a second main module for performing local region segmentation on the acquired sheep body point cloud data from the top, left, and right views to obtain overlapping point cloud regions; a third main module for using the point cloud data from the left and right views as source point clouds and the point cloud data from the top view as target point clouds, and introducing an improved CPD algorithm to register the target point cloud and source point cloud in the overlapping point cloud regions to obtain local point cloud rigidity changes in the left and right views; wherein, the improved CPD algorithm introduces chamfer distance judgment; a fourth main module for performing rotation matrix and translation vector transformations on the complete point clouds of the left and right views based on the local point cloud rigidity changes in the left and right views to obtain complete point cloud rigidity changes in the left and right views; and a fifth main module for merging the top view with the transformed left and right views to obtain the registered three-dimensional sheep body point cloud.

[0090] This invention provides a sheep body 3D point cloud registration system based on locally overlapping regions. Addressing the problems of low accuracy and slow registration speed in existing point cloud registration, this system employs the aforementioned modules and avoids the low accuracy of overall registration of point clouds without significant features by using registration of overlapping regions. Furthermore, due to the local registration, it has a faster registration speed compared to global registration.

[0091] It should be noted that the system embodiments provided by this invention, in addition to implementing the methods in the above method embodiments, are also used to implement the methods in other method embodiments provided by this invention. The difference lies only in setting corresponding functional modules, and their principles are basically the same as those of the above system embodiments provided by this invention. As long as those skilled in the art, based on the above system embodiments and referring to the specific technical solutions in other method embodiments, obtain corresponding technical means and technical solutions composed of these technical means by combining technical features, and improve the modules in the above system embodiments while ensuring the practicality of the technical solutions, they can obtain corresponding system-like embodiments for implementing the methods in other method-like embodiments. For example:

[0092] Based on the above system embodiments, as a preferred embodiment, the sheep body 3D point cloud registration system based on locally overlapping regions provided in this embodiment of the invention further includes:

[0093] The downsampling module is used to downsample the sheep body point cloud data from the top view, left view, and right view, respectively.

[0094] Based on the above system embodiments, as a preferred embodiment, the sheep body 3D point cloud registration system based on locally overlapping regions provided in this embodiment of the invention further includes:

[0095] The point cloud segmentation module is used to estimate possible overlapping areas of point clouds using the three-dimensional perspective of multiple cameras, and to perform local point cloud segmentation based on the estimation results to obtain the overlapping areas of point clouds.

[0096] Based on the above system embodiments, as a preferred embodiment, this invention provides a sheep body 3D point cloud registration system based on locally overlapping regions, wherein the third main module is further configured to execute the following instructions:

[0097] The source point cloud in the left or right view is subjected to iterative rigid changes, and convergence is judged when the number of iterations has not reached the specified value.

[0098] When the algorithm convergence between the target point cloud and the source point cloud fails to reach the rated value, the chamfer distance is judged, and the rigidity change is adjusted for the next iteration.

[0099] If the chamfer distance between the target point cloud and the source point cloud does not reach the specified value, proceed to the next iteration;

[0100] When the number of iterations reaches the specified value, the algorithm converges to the specified value, or the chamfer distance reaches the specified value, the iteration exits and the local rigidity change of the left or right view is output.

[0101] Based on the above system embodiments, as a preferred embodiment, this invention provides a sheep body 3D point cloud registration system based on locally overlapping regions, wherein the third main module is further configured to execute the following instructions:

[0102] Calculate the distance matrix between the target point cloud and the source point cloud after iterative changes;

[0103] The calculated distance matrix is ​​converted into a probability matrix and then normalized.

[0104] Adjust the rigidity change matrix based on the normalized probability matrix;

[0105] Based on the adjusted rigidity change matrix and the calculated auxiliary variables, the covariance of the Gaussian mixture model is updated, and the algorithm convergence value is determined based on the absolute difference between the covariance before and after the update.

[0106] Based on the above system embodiments, as a preferred embodiment, this invention provides a sheep body 3D point cloud registration system based on locally overlapping regions, wherein the third main module is further configured to execute the following instructions:

[0107] Normalize the target point cloud and the source point cloud after rigid changes;

[0108] Based on the normalization results, determine whether the chamfer distance between the target point cloud and the source point cloud reaches the specified value. The chamfer distance is calculated as follows:

[0109]

[0110] Where X and Y' represent the target point cloud and the source point cloud after rigid transformation, respectively, and ||.||2 represents the Euclidean distance.

[0111] Based on the same inventive concept as the above embodiments, this embodiment of the invention also provides a method and system for registering three-dimensional point clouds of sheep bodies based on locally overlapping regions, including a memory and a processor. The memory stores program instructions that are executed by the processor, and the processor calls the program instructions to perform the steps of the method for registering three-dimensional point clouds of sheep bodies based on locally overlapping regions.

[0112] Based on the same inventive concept as the above embodiments, this embodiment of the invention also provides a non-transitory computer-readable storage medium storing computer instructions that cause the computer to execute the steps of the sheep body three-dimensional point cloud registration method based on locally overlapping regions.

[0113] In summary, this invention uniformly downsamples the filtered point cloud data, estimates the proportional relationship between the overlapping areas of the target detection bodies covered by the vertical and horizontal cameras and the overall point cloud using the camera's field of view, introduces chamfer distance judgment into the CPD algorithm, obtains the transformation matrix of the local overlapping area, applies the local transformation matrix to perform corresponding rotation matrix and translation vector transformations on the left and right complete point clouds respectively, and finally merges the transformed left, right, and top view point clouds to obtain the registered 3D sheep body point cloud. This invention effectively solves the problem of registration failure caused by getting trapped in local optima when initializing position sensitive and asymmetric geometry. Furthermore, by introducing chamfer distance, the registration speed is increased by 3 times compared to the previous CPD algorithm.

[0114] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features therein; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the technical solutions of the embodiments of the present invention.

Claims

1. A method for registering 3D point clouds of sheep bodies based on locally overlapping regions, characterized in that, include: Obtain the point cloud data of the sheep body from the top view, left view, and right view respectively; The point cloud data of the sheep body in the top view, left view and right view are segmented locally to obtain the overlapping area of ​​the point cloud; Using point cloud data from the left and right views as source point clouds and point cloud data from the top view as target point clouds, an improved CPD algorithm is introduced to register the target point cloud and source point cloud in the overlapping area of ​​the point clouds, resulting in local point cloud rigidity changes in the left and right views. The improved CPD algorithm incorporates a chamfer distance judgment. The registration of the target point cloud and source point cloud in the overlapping area using the improved CPD algorithm includes: iteratively performing rigidity changes on the source point cloud of the left or right view; performing a convergence judgment if the number of iterations has not reached a predetermined value; performing a chamfer distance judgment if the algorithm convergence between the target and source point clouds has not reached a predetermined value, and adjusting the rigidity changes for the next iteration; proceeding to the next iteration if the chamfer distance between the target and source point clouds has not reached a predetermined value; and exiting the iteration when the number of iterations reaches a predetermined value, the algorithm converges to a predetermined value, or the chamfer distance reaches a predetermined value, outputting the local rigidity changes of the left or right view. Based on the rigidity changes of the local point clouds in the left and right views, the complete point clouds in the left and right views are transformed by rotation matrix and translation vector to obtain the rigidity changes of the complete point clouds in the left and right views. The top view is merged with the transformed left and right views to obtain the registered 3D sheep point cloud.

2. The sheep body 3D point cloud registration method based on locally overlapping regions according to claim 1, characterized in that, After obtaining the sheep body point cloud data from the top, left, and right views, the following is also included: The point cloud data of the sheep body in the top view, left view, and right view were downsampled respectively.

3. The sheep body 3D point cloud registration method based on locally overlapping regions according to claim 2, characterized in that, After downsampling the sheep body point cloud data from the top, left, and right views, the following is also included: Utilize the 3D perspective of multiple cameras to estimate possible overlapping areas of point clouds; Based on the budget results, local point cloud segmentation is performed to obtain the overlapping area of ​​the point cloud.

4. The sheep body 3D point cloud registration method based on locally overlapping regions according to claim 1, characterized in that, Perform chamfer distance determination, including: Normalize the target point cloud and the source point cloud after rigid changes; Based on the normalization results, determine whether the chamfer distance between the target point cloud and the source point cloud reaches the specified value. The chamfer distance is calculated as follows: ， Where X and Y' represent the target point cloud and the source point cloud after rigid transformation, respectively, and ||.||2 represents the Euclidean distance.

5. A three-dimensional point cloud registration system for sheep bodies based on locally overlapping regions, characterized in that, include: The first main module is used to acquire point cloud data of the sheep body from the top view, left view, and right view, respectively. The second main module is used to perform local region segmentation on the acquired top view, left view and right view sheep body point cloud data respectively to obtain the point cloud overlapping area; The third main module uses point cloud data from the left and right views as source point clouds and point cloud data from the top view as target point clouds. It introduces an improved CPD algorithm to register the target and source point clouds in overlapping areas, obtaining local point cloud rigidity changes in the left and right views. The improved CPD algorithm incorporates a chamfer distance judgment. The registration of the target and source point clouds in overlapping areas using the improved CPD algorithm includes: iteratively changing the source point cloud in either the left or right view; performing a convergence judgment if the number of iterations has not reached a predetermined value; judging the chamfer distance if the algorithm convergence between the target and source point clouds has not reached a predetermined value, and adjusting the rigidity change for the next iteration; proceeding to the next iteration if the chamfer distance between the target and source point clouds has not reached a predetermined value; and exiting the iteration when the number of iterations, algorithm convergence, or chamfer distance reaches a predetermined value, outputting the local rigidity changes in the left or right view. The fourth main module is used to perform rotation matrix and translation vector transformations on the complete point clouds of the left and right views based on the local point cloud rigidity changes of the left and right views, so as to obtain the complete point cloud rigidity changes of the left and right views. The fifth main module is used to merge the top view with the transformed left and right views to obtain the registered 3D sheep point cloud.

6. A sheep body 3D point cloud registration device based on locally overlapping regions, characterized in that, The device includes a memory and a processor, wherein the memory stores program instructions that are executed by the processor, and the processor invokes the program instructions to perform the steps of the sheep body three-dimensional point cloud registration method based on local overlapping regions as described in any one of claims 1 to 4.

7. A non-transitory computer-readable storage medium, characterized in that, The non-transitory computer-readable storage medium stores computer instructions that cause the computer to perform the steps of the sheep body three-dimensional point cloud registration method based on locally overlapping regions as described in any one of claims 1 to 4.

Images