A Candidate Region Generation Method and System

Abstract

The present invention discloses a candidate region generation method and system. Among them, the method includes: obtaining point cloud input data, and voxelizing the point cloud input data to obtain voxel data; extracting feature data according to the voxel data; generating region similarity according to the feature data; and merging discrete regions according to the region similarity to obtain candidate regions. The present invention solves the technical problem that the prior art method based on sliding cubes usually assumes that the ground objects are framed inside the cubes, and the treatment effect for the adhesion phenomenon of the ground objects is poor.

Description

Technical Field

[0001] The present invention relates to the field of target recognition, and more particularly, to a method and system for generating candidate regions. Background Art

[0002] Three-dimensional target recognition technology (3D Target Recognition Technology) can quickly, automatically, and accurately collect a large amount of three-dimensional point data on the surface of target objects in virtual reality (VR) and augmented reality (AR), namely, build point clouds (Point Clouds). It targets the input large scene, performs ground point cloud segmentation according to the already trained database, and finally identifies and models the objects in the large scene.

[0003] Three-dimensional target recognition technology is widely used in computer vision and virtual reality, such as the establishment of indoor three-dimensional spaces, the rapid modeling of outdoor buildings, the construction of three-dimensional virtual sand tables, and the establishment of three-dimensional virtual cities. Compared with the previous two-dimensional recognition, three-dimensional recognition can present the results more intuitively and accurately, with higher visibility and ornamental value.

[0004] However, after the ground point cloud is segmented, non-ground points often adhere to each other, especially the overlap between different ground objects (such as buildings and trees). As a result, the final presented result cannot individualize each object, and two or more objects are regarded as one object. The traditional method is to directly use prior information or geometric feature matching methods, which requires constructing a sliding cube, extracting local features or feature bags from the point cloud within the cube, and finally classifying the point cloud through a classifier. Such methods are almost an exhaustive search of the point cloud, and also need to consider the influence of scale and rotation angle, and the overall calculation is very time-consuming. In addition, the method based on the sliding cube usually assumes that the ground object is enclosed within the cube, and the treatment effect of the adhesion phenomenon of the ground object is poor, and the final presented result is not good.

[0005] After the ground point cloud is segmented by the traditional fast candidate target detection method, non-ground points often adhere to each other, especially the overlap between different ground objects (such as buildings and trees). If directly using prior information or geometric feature matching methods, a sliding cube needs to be constructed, local features or feature bags are extracted from the point cloud within the cube, and finally the point cloud is classified through a classifier. Such methods are almost an exhaustive search of the point cloud, and also need to consider the influence of scale and rotation angle, and the calculation is very time-consuming. In addition, the method based on the sliding cube usually assumes that the ground object is enclosed within the cube, and the treatment effect of the adhesion phenomenon of the ground object is poor.

[0006] In view of the above problems, no effective solution has been proposed yet. Summary of the Invention

[0007] An embodiment of the present invention provides a method and system for generating candidate regions to at least solve the technical problem that the prior art method based on a sliding cube usually assumes that the ground object is enclosed inside the cube and has a poor processing effect on the adhesion phenomenon of the ground object.

[0008] According to one aspect of the embodiments of the present invention, a method for generating candidate regions is provided, including: obtaining point cloud input data, and voxelizing the point cloud input data to obtain voxel data; extracting feature data according to the voxel data; generating a region similarity according to the feature data; and merging discrete regions according to the region similarity to obtain candidate regions.

[0009] Optionally, the extracting feature data according to the voxel data includes: generating a color similarity according to the voxel data; and extracting the feature data according to the color similarity.

[0010] Optionally, before generating the region similarity according to the feature data, the method further includes: obtaining a feature frequency according to the feature data; and obtaining a texture similarity and a morphological similarity according to the feature frequency.

[0011] Optionally, the candidate regions include: a segmented point cloud scene.

[0012] According to another aspect of the embodiments of the present invention, a system for generating candidate regions is further provided, including: an obtaining module, configured to obtain point cloud input data, and voxelize the point cloud input data to obtain voxel data; an extracting module, configured to extract feature data according to the voxel data; a generating module, configured to generate a region similarity according to the feature data; and a merging module, configured to merge discrete regions according to the region similarity to obtain candidate regions.

[0013] Optionally, the extracting module includes: a generating unit, configured to generate a color similarity according to the voxel data; and an extracting unit, configured to extract the feature data according to the color similarity.

[0014] Optionally, the system further includes: a frequency module, configured to obtain a feature frequency according to the feature data; and a similarity module, configured to obtain a texture similarity and a morphological similarity according to the feature frequency.

[0015] Optionally, the candidate regions include: a segmented point cloud scene.

[0016] According to another aspect of the embodiments of the present invention, a non-volatile storage medium is further provided, where the non-volatile storage medium includes a stored program, and when the program runs, it controls a device where the non-volatile storage medium is located to execute a method for generating candidate regions.

[0017] According to another aspect of the embodiments of the present invention, an electronic device is further provided, which includes a processor and a memory; computer-readable instructions are stored in the memory, and the processor is configured to run the computer-readable instructions, wherein when the computer-readable instructions run, a candidate region generation method is executed.

[0018] In the embodiments of the present invention, by obtaining point cloud input data, voxelizing the point cloud input data to obtain voxel data; extracting feature data according to the voxel data; generating region similarity according to the feature data; and merging discrete regions according to the region similarity to obtain candidate regions, the technical problem that the prior art method based on a sliding cube usually assumes that the ground object is enclosed inside the cube and has a poor processing effect on the adhesion phenomenon of the ground object is solved. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] The drawings described herein are used to provide a further understanding of the present invention and constitute a part of this application. The schematic embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute an improper limitation to the present invention. In the drawings:

[0020] Figure 1 is a flowchart of a candidate region generation method according to an embodiment of the present invention;

[0021] Figure 2 is a structural block diagram of a candidate region generation system according to an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0022] In order to enable those skilled in the art to better understand the solutions of the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0023] It should be noted that the terms "first", "second", etc. in the specification, claims and above-mentioned drawings of the present invention are used to distinguish similar objects, and do not necessarily describe a specific order or sequence. It should be understood that the data used in this way can be interchanged under appropriate circumstances, so that the embodiments of the present invention described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "comprising" and "having" and any variations thereof are intended to cover non-exclusive inclusion. For example, a process, method, system, product or device that includes a series of steps or units does not necessarily limit to those steps or units clearly listed, but may include other steps or units not clearly listed or inherent to these processes, methods, products or devices.

[0024] According to an embodiment of the present invention, a method embodiment of a candidate region generation method is provided. It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions. And although the logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in an order different from that here.

[0025] Embodiment 1

[0026] Figure 1 is a flowchart of a candidate region generation method according to an embodiment of the present invention, as Figure 1 shown, the method includes the following steps:

[0027] Step S102, obtain point cloud input data, and voxelize the point cloud input data to obtain voxel data.

[0028] Step S104, extract feature data according to the voxel data.

[0029] Step S106, generate region similarity according to the feature data.

[0030] Step S108, merge discrete regions according to the region similarity to obtain candidate regions.

[0031] Optionally, the extracting feature data according to the voxel data includes: generating color similarity according to the voxel data; extracting the feature data according to the color similarity.

[0032] Optionally, before generating the region similarity according to the feature data, the method further includes: obtaining feature frequency according to the feature data; obtaining texture similarity and morphological similarity according to the feature frequency.

[0033] Optionally, the candidate region includes: a segmented point cloud scene.

[0034] Specifically, the embodiments of the present invention belong to the field of image processing. In recent years, great progress has been made in the research on rapid detection of candidate targets. Especially the Selective Search (SS) strategy makes full use of the texture distribution of the target. Through rapid region feature extraction and analysis, the search range of candidate targets can be greatly reduced. Therefore, based on non-ground point cloud segmentation, the present invention designs a rapid detection method for candidate ground objects according to the characteristics of point cloud distribution as the input of the deep learning model, which can make the selected objects of different ground objects more robust, and at the same time have consistency and efficiency.

[0035] In addition, the rapid generation algorithm for ground object candidate regions based on the hierarchical merging tree provided by the embodiments of the present invention includes the following steps:

[0036] Step 101: Use an octree to divide the input point cloud to achieve voxelization, and then evenly arrange region growth nuclei in the voxel space.

[0037] Step 102: Incorporate the square with the minimum distance function value of the point cloud supervoxel into the nucleus, and finally all nuclei complete growth almost simultaneously.

[0038] Step 103: Based on Step 102, the original point cloud can be divided into multiple different regions, and it is ensured that the nuclei in each region are similar.

[0039] Step 104: Calculate the R, G, B color histograms of all corresponding positions of every two adjacent point cloud regions respectively. The horizontal axis is the color intensity values of R, G, B, and the vertical axis is the proportion of the intensity value in the whole.

[0040] Step 105: Calculate the L2 norm of the corresponding color intensity values of every two adjacent point cloud regions in the R, G, B color histograms to obtain the color similarity, that is

[0041] Step 106: Extract FPFH features from the large-scale point cloud. For the features of each point and its 8 neighbor points, the maximum expectation is calculated using the Gaussian mixture model. And the obtained 9 feature sequences are used as the feature codebook of the point.

[0042] Step 107: Count the frequency of each feature in the codebook to obtain the morphological intensity. Then calculate the average value of the vectors with small intensity differences of the same category features in two regions, and further obtain the morphological similarity.

[0043] Step 108: Extract the LBP texture features of each point in the candidate region. Calculate the k nearest neighbor points for each point. If the pixel of the central point is higher than the neighbor points, the neighbor points are assigned 1, otherwise assigned 0.

[0044] Step 109: Each point has a string of binary features of length k. Furthermore, the intensity information of each point can be counted. By taking the difference of the sum of the intensity information of the candidate regions, the texture similarity S can be obtained. t (R i , R j ).

[0045] Step 110: Based on Steps 104 - 109, multiply the color similarity S c , morphological similarity S s , and texture similarity S t of the two regions by the weight coefficients α, β, and λ respectively to obtain the regional comprehensive similarity: S(R i , R j ) = αS c (R i , R j ) + βS s (R i , R j ) + γS t (R i , R j ).

[0046] Step 111: Find the two most similar regions according to the regional comprehensive similarity, merge them, calculate the histogram of the merged region to obtain a new vector. Calculation method: Calculate the norms of the two old regions respectively, denoted as ||x i || 2 and ||x j || 2 , then multiply them by the vectors and respectively. Finally, add them up and divide by the value of the norm of the two regions as a whole to obtain the result

[0047] Step 112: Delete the old regions and the corresponding similarities, and recalculate the similarities of the newly generated regions and their adjacent regions after merging.

[0048] Step 113: Calculate the supervoxel distance value and the reciprocal of the regional comprehensive similarity for each region respectively, and perform a weighted sum of the two to obtain the response value of each region.

[0049] Step 114: Based on all the response values globally, use the greedy strategy to construct multiple minimum spanning trees, thereby obtaining multiple minimum weight paths.

[0050] Step 115: Merge the two smallest paths to achieve the merging of the two most similar region blocks.

[0051] Step 116: Repeatedly execute Steps 104 - 115 for all regions to be merged, which can achieve the merging of discrete regions and finally form the segmented point cloud scene.

[0052] In order to better achieve single - object segmentation of objects in a large - scale scene, this project plans to design a fast detection method for candidate ground objects based on the characteristics of point cloud distribution on the basis of non - ground point cloud segmentation, so as to be used as the input of a deep - learning model to classify the targets in virtual reality. Compared with two - dimensional images, the shapes of ground objects in three - dimensional point clouds vary greatly, and the point cloud distribution shows irregular characteristics. Therefore, the method for selecting candidate ground objects must meet the following three characteristics: (1) Scale adaptability. That is, it should be robust to ground objects of different sizes; (2) Consistency. The internal color, texture, and shape differences of the candidate ground object point cloud should not be too large; (3) Efficiency. The point cloud data volume is very large, and the candidate set is only the basis for subsequent classification and should not consume too much computing resources.

[0053] It should be noted that curvature filtering is an optimization algorithm in image processing, which first appeared in Chapter 6 of Dr. Gong Yuanhao's doctoral thesis (ETH E - Collection: Spectrally regularized surfaces). Whether it is the denoising and smoothing problems in two - dimensional images or in three - dimensional point clouds, they are usually ill - posed problems, and ill - posed problems require regularization terms. Curvature regularization is a commonly used regularization term for ill - posed problems, and the obtained models are usually good, but these models are also difficult to solve. There are two traditional solution methods: one is based on the gradient descent method (diffusion formula), and the other is based on the Euler - Lagrange equation. Usually, the latter solution method is more efficient than the former, but how to obtain this equation is usually very complex, and it is difficult to see its corresponding physical meaning from the obtained equation. Curvature filtering considers this optimization problem from another perspective. It is a kind of filtering, but it also optimizes a certain regularization term, and implicitly uses known surfaces in differential geometry during the filtering process, so there is no need to calculate the Gaussian curvature or the mean curvature, reducing the computational complexity. The advantages of curvature filtering are: high efficiency, which is 100 to 1000 times faster than traditional methods; generality, which can solve any complex noise model; theoretical guarantee, based on classical differential set theory; easy to implement and parallel.

[0054] The "Fast Generation Algorithm for Ground Object Candidate Regions Based on Hierarchical Merge Trees" of the present invention is an innovation and generalization based on the idea of curvature filtering. The point cloud distortion correction method of the present invention does not need to optimize the regularization term or optimize based on the Euler - Lagrange equation. The present invention only needs to update each point of the point cloud data according to the specified rules, which can reduce the computational complexity and improve the effect and efficiency of smoothing the point cloud data.

[0055] Through the above embodiments, the technical problem that the prior art method based on the sliding cube usually assumes that the ground objects are framed inside the cube and has a poor processing effect on the adhesion phenomenon of the ground objects is solved.

[0056] Embodiment 2

[0057] Figure 2 is a structural block diagram of a candidate region generation system according to an embodiment of the present invention, as Figure 2 shown, the system includes:

[0058] An acquisition module 20, configured to acquire point cloud input data, and perform voxelization on the point cloud input data to obtain voxel data.

[0059] An extraction module 22, configured to extract feature data according to the voxel data.

[0060] A generation module 24, configured to generate a region similarity according to the feature data.

[0061] A merging module 26, configured to merge discrete regions according to the region similarity to obtain candidate regions.

[0062] Optionally, the extraction module includes: a generation unit, configured to generate a color similarity according to the voxel data; an extraction unit, configured to extract the feature data according to the color similarity.

[0063] Optionally, the system further includes: a frequency module, configured to obtain a feature frequency according to the feature data; a similarity module, configured to obtain a texture similarity and a morphological similarity according to the feature frequency.

[0064] Optionally, the candidate region includes: a segmented point cloud scene.

[0065] Specifically, the embodiment of the present invention belongs to the field of image processing. In recent years, great progress has been made in the research on rapid detection of candidate targets. Especially the selective search strategy (SS) makes full use of the texture distribution of the targets and can greatly reduce the search range of candidate targets through rapid region feature extraction and analysis. Therefore, based on the non-ground point cloud segmentation, the present invention designs a rapid detection method for candidate ground objects according to the characteristics of the point cloud distribution as the input of the deep learning model. It can make the selected objects of different ground objects more robust, and at the same time have consistency and high efficiency.

[0066] In addition, the rapid generation algorithm for ground object candidate regions based on the hierarchical merge tree provided by the embodiment of the present invention includes the following steps:

[0067] Step 101: Use an octree to divide the input point cloud to achieve voxelization, and then evenly arrange region growth nuclei in the voxel space.

[0068] Step 102: Incorporate the cube with the minimum distance function value of the point cloud supervoxel into the crystal nucleus, and finally all crystal nuclei complete growth almost simultaneously.

[0069] Step 103: Based on Step 102, the original point cloud can be divided into multiple different regions, and it is ensured that the crystal nuclei in each region are similar.

[0070] Step 104: Calculate the R, G, and B color histograms of all corresponding positions of every two adjacent point cloud regions respectively. The horizontal axis is the color intensity values of R, G, and B, and the vertical axis is the proportion of this intensity value in the whole.

[0071] Step 105: Calculate the L2 norm of the corresponding color intensity values of every two adjacent point cloud regions in the R, G, and B color histograms to obtain the color similarity, that is

[0072] Step 106: Extract FPFH features from the large-scale point cloud. For the features of each point and its 8 neighbor points, the maximum expectation is calculated using the Gaussian mixture model. And the obtained 9 feature sequences are used as the feature codebook of this point.

[0073] Step 107: Count the frequency of each feature in the codebook to obtain the morphological intensity. Then calculate the average value of the vectors with small intensity differences of the same category features in two regions, and further obtain the morphological similarity.

[0074] Step 108: Extract the LBP texture features of each point in the candidate region. Calculate the k nearest neighbor points for each point. If the pixel of the central point is higher than the neighbor points, the neighbor points are assigned 1, otherwise 0.

[0075] Step 109: Each point has a string of binary features with a length of k. Furthermore, the intensity information of each point can be counted. Calculate the difference of the sum of the intensity information of the candidate region to obtain the texture similarity S t (R i , R j ).

[0076] Step 110: Based on Steps 104 - 109, multiply the color similarity S c , the morphological similarity S s and the texture similarity S t by the weight coefficients α, β, and λ respectively to obtain the regional comprehensive similarity: S(R i , R j ) = αS c (R i , R j ) + βS s (R i , Rj ) + γS t (R i ,R j )。

[0077] Step 111: Find the two most similar regions according to the regional comprehensive similarity, merge them, calculate the histogram of the merged region, and obtain a new vector. Calculation method: Calculate the norms of the two old regions respectively, denoted as ||x i || 2 and ||x j || 2 , then multiply them by the vectors and Finally, add them up and divide by the norm value of the overall two regions to get the result

[0078] Step 112: Delete the old regions and their corresponding similarities, and recalculate the similarities of the newly generated regions and their adjacent regions after merging.

[0079] Step 113: Calculate the supervoxel distance value and the reciprocal of the regional comprehensive similarity for each region respectively, and perform a weighted sum of the two to obtain the response value of each region.

[0080] Step 114: Based on all the response values globally, use the greedy strategy to construct multiple minimum spanning trees, so as to obtain multiple minimum weight paths.

[0081] Step 115: Merge the two smallest paths to achieve the merging of the two most similar region blocks.

[0082] Step 116: Repeatedly execute Steps 104 - 115 for all regions to be merged to achieve the merging of discrete regions, and finally form the segmented point cloud scene.

[0083] In order to better achieve single - object segmentation of objects in large scenes, this project plans to design a fast detection method for candidate ground objects based on the characteristics of point cloud distribution on the basis of non - ground point cloud segmentation, so as to be used as the input of the deep - learning model to classify the targets in virtual reality. Compared with two - dimensional images, the ground object shapes in three - dimensional point clouds vary greatly, and the point cloud distribution shows irregular characteristics. Therefore, the method for selecting candidate ground objects must meet the following three characteristics: (1) Scale adaptability. That is, it should be robust to ground objects of different sizes; (2) Consistency. The internal color, texture, and morphological differences of the candidate ground object point cloud should not be too large; (3) Efficiency. The point cloud data volume is very large, and the candidate set is only the basis for subsequent classification and should not consume too much computing resources.

[0084] It should be noted that curvature filtering is an optimization algorithm in image processing, which first appeared in Chapter 6 of Dr. Gong Yuanhao's doctoral thesis (ETH E-Collection: Spectrally regularized surfaces). Whether it is the denoising and smoothing problems in 2D images or in 3D point clouds, they are usually ill-posed, and ill-posed problems require regularization terms. Curvature regularization is a commonly used regularization term for ill-posed problems, and the resulting models are usually good, but these models are also difficult to solve. There are two traditional solution methods: one is based on the gradient descent method (diffusion formula), and the other is based on the Euler-Lagrange equation. Usually, the latter solution method is more efficient than the former, but how to obtain this equation is usually very complicated, and it is difficult to see its corresponding physical meaning from the obtained equation. Curvature filtering, on the other hand, considers this optimization problem from another perspective. It is a kind of filtering, but it also optimizes a certain regularization term, and in the process of filtering, it implicitly uses the known surfaces of differential geometry, so there is no need to calculate the Gaussian curvature or the mean curvature, reducing the computational complexity. The advantages of curvature filtering are: high efficiency, 100 to 1000 times faster than traditional methods; generality, it can solve any complex noise model; theoretical guarantee, based on classical differential set theory; easy to implement and parallelize.

[0085] The "Fast Generation Algorithm for Feature Candidate Regions Based on Hierarchical Merge Trees" of the present invention is an innovation and generalization based on the idea of curvature filtering. The point cloud distortion correction method of the present invention does not need to optimize the regularization term or optimize based on the Euler-Lagrange equation. The present invention only needs to update each point of the point cloud data according to the specified rules, which can reduce the computational complexity and improve the effect and efficiency of smoothing the point cloud data.

[0086] According to another aspect of the embodiments of the present invention, a non-volatile storage medium is also provided. The non-volatile storage medium includes a stored program, wherein when the program runs, it controls the device where the non-volatile storage medium is located to execute a candidate region generation method.

[0087] According to another aspect of the embodiments of the present invention, an electronic device is also provided, including a processor and a memory; computer-readable instructions are stored in the memory, and the processor is used to run the computer-readable instructions, wherein when the computer-readable instructions run, they execute a candidate region generation method.

[0088] Through the above embodiments, the technical problem that the existing method based on the sliding cube usually assumes that the feature is enclosed inside the cube and has a poor processing effect on the adhesion phenomenon of the feature is solved.

[0089] The serial numbers of the above embodiments of the present invention are only for description and do not represent the advantages and disadvantages of the embodiments.

[0090] In the above embodiments of the present invention, the descriptions of the various embodiments each have their own emphasis. For the parts not detailed in a certain embodiment, reference may be made to the relevant descriptions of other embodiments.

[0091] In the several embodiments provided by the present application, it should be understood that the disclosed technical content can be implemented in other ways. Among them, the device embodiments described above are only illustrative. For example, the division of the units can be a logical function division. In actual implementation, there can be other division methods. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the displayed or discussed coupling or direct coupling or communication connection between each other can be through some interfaces. The indirect coupling or communication connection of units or modules can be in an electrical or other form.

[0092] The units described as separate components may or may not be physically separated. The components displayed as units may or may not be physical units, that is, they can be located in one place or distributed to multiple units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0093] In addition, in each embodiment of the present invention, the functional units can be integrated in a processing unit, or each unit can exist physically alone, or two or more units can be integrated in one unit. The above-mentioned integrated units can be implemented in the form of hardware or in the form of software functional units.

[0094] If the above-mentioned integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of this technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in each embodiment of the present invention. And the aforementioned storage medium includes: USB flash drives, read-only memories (ROM, Read-Only Memory), random access memories (RAM, Random Access Memory), mobile hard disks, magnetic disks, or optical disks, etc., which can store program codes.

[0095] The above are only the preferred embodiments of the present invention. It should be noted that for those of ordinary skill in the art, without departing from the principle of the present invention, several improvements and refinements can be made, and these improvements and refinements should also be regarded as the protection scope of the present invention.

Claims

1. A candidate region generation method, characterized in that Including: Obtain point cloud input data, and voxelize the point cloud input data to obtain voxel data; Generate color similarity according to the voxel data, extract feature data according to the color similarity, obtain feature frequency according to the feature data, and obtain texture similarity and morphological similarity according to the feature frequency; Generate regional comprehensive similarity by multiplying the color similarity, texture similarity, and morphological similarity by weight coefficients; Find the two most similar regions according to the regional comprehensive similarity, merge them, delete the old regions and the corresponding similarities, and recalculate the similarities of the newly generated regions and their adjacent regions after merging. Calculate the supervoxel distance value and the reciprocal of the regional comprehensive similarity for each region respectively, and perform weighted summation on the two to obtain the response value of each region. Based on all the response values, start from the global perspective and use the greedy strategy to construct multiple minimum spanning trees, so as to obtain multiple minimum weight paths. Merge the two smallest paths to achieve the merging of the two most similar region blocks and obtain candidate regions.

2. The method according to claim 1, wherein The candidate regions include: the segmented point cloud scene.

3. A candidate region generation system, characterized in that Including: An acquisition module for obtaining point cloud input data and voxelizing the point cloud input data to obtain voxel data; An extraction module for generating color similarity according to the voxel data, performing FPFH feature extraction on large-scale point cloud data to extract feature data; obtaining feature frequency according to the feature data, and obtaining texture similarity and morphological similarity according to the feature frequency; A generation module for generating regional comprehensive similarity by multiplying the color similarity, texture similarity, and morphological similarity by weight coefficients; A merging module for finding the two most similar regions according to the regional comprehensive similarity, merging them, deleting the old regions and the corresponding similarities, and recalculating the similarities of the newly generated regions and their adjacent regions after merging. Calculate the supervoxel distance value and the reciprocal of the regional comprehensive similarity for each region respectively, and perform weighted summation on the two to obtain the response value of each region. Based on all the response values, start from the global perspective and use the greedy strategy to construct multiple minimum spanning trees, so as to obtain multiple minimum weight paths. Merge the two smallest paths to achieve the merging of the two most similar region blocks and obtain candidate regions.

4. The system according to claim 3, characterized in that, The candidate regions include: the segmented point cloud scene.

5. A non-volatile storage medium, characterized in that, The non-volatile storage medium includes a stored program, wherein the program, when running, controls the device where the non-volatile storage medium is located to execute the method according to any one of claims 1 to 2.

6. An electronic device, characterized in that, Comprising a processor and a memory; computer-readable instructions are stored in the memory, and the processor is used to run the computer-readable instructions, wherein the computer-readable instructions, when running, execute the method according to any one of claims 1 to 2.

Images