Point cloud data segmentation method based on structure information theory, terminal and storage medium

Abstract

The invention provides a point cloud data segmentation method based on a structure information theory, a terminal and a storage medium. The method comprises the following steps: projecting point cloud data to a two-dimensional plane from different dimensions to obtain two-dimensional images; for each two-dimensional image, based on the distance between pixel points in the two-dimensional image, creating an edge by taking the maximum one-dimensional structure entropy after the edge is created for the two-dimensional image as a target, and obtaining an interaction graph corresponding to the two-dimensional image; for each interaction image, based on the expected number of superpixels and the weight of each edge in the interaction image, performing image division by taking the minimum entropy of a high-dimensional structure after the pixels in the interaction image are combined as a target to obtain a division result of the interaction image; obtaining a superpixel segmentation result of each two-dimensional image according to the division result of each interaction image; and performing inverse projection on the superpixel segmentation result of each two-dimensional image to obtain a segmentation result of the point cloud data. According to the method, higher segmentation precision, better adaptability and a more stable segmentation result can be provided.

Description

Technical Field

[0001] The present invention relates to the field of data processing technology, and in particular to a point cloud data segmentation method, a terminal and a storage medium based on structural information theory. Background Art

[0002] Point cloud data segmentation is an important task in computer vision and 3D data processing, especially in the fields of autonomous driving, 3D reconstruction, virtual reality and robot navigation. The goal of point cloud segmentation is to divide point cloud data into several sub-regions with semantic consistency or geometric similarity. Through segmentation, different objects or surface areas (such as cars, pedestrians, roads, etc.) can be identified to provide support for subsequent 3D recognition and reconstruction tasks.

[0003] Since point cloud data is sparse and disordered, it is very complicated to process the geometric relationship between points and the neighborhood calculation. The superpixel segmentation method of two-dimensional images can simplify the calculation through the regularity of the image structure. At the same time, the pixel grid of the two-dimensional image is more intuitive and efficient when processing the relationship between adjacent pixels. The three-dimensional point cloud segmentation method usually only relies on geometric structure information, and it is difficult to comprehensively utilize multiple modal data to improve the segmentation results. After mapping the point cloud to a two-dimensional image, superpixel segmentation can more easily combine color information. Through two-dimensional projection, the uneven point cloud data can be mapped to a regular pixel grid, thereby improving the accuracy of segmentation.

[0004] However, the inventors found that traditional superpixel segmentation methods such as Simple Linear Iterative Clustering (SLIC) and Superpixels Extracted via Energy-Driven Sampling (SEEDS) usually only consider the local similarity between pixels and rely on adjacent pixels or local features of pixels, resulting in unstable segmentation results. In traditional methods, the number or scale of superpixels needs to be set manually and lacks scalability and flexibility, which will lead to under-segmentation or over-segmentation, making it difficult to combine with other advanced technologies for further optimization. When processing larger images or complex structures, traditional methods will have incoherent segmentation results or incomplete topological structures. Many superpixel segmentation methods based on clustering or energy optimization are prone to errors at complex boundaries, cannot capture the true contours of objects well, are unstable in different application scenarios, and rely on specific features of the image. Summary of the invention

[0005] The embodiments of the present invention provide a point cloud data segmentation method, a terminal and a storage medium based on structural information theory to solve the problems that the current point cloud data segmentation method has unstable performance and is inaccurate when the topological structure is complex.

[0006] In a first aspect, an embodiment of the present invention provides a point cloud data segmentation method based on structural information theory, comprising: Project the point cloud data from different dimensions onto a two-dimensional plane to obtain individual two-dimensional images; For each two-dimensional image, based on the distance between pixels in the two-dimensional image, an edge is created with the goal of maximizing the one-dimensional structural entropy after the edge is created for the two-dimensional image, and an interaction graph corresponding to the two-dimensional image is obtained; For each interaction graph, based on the expected number of superpixels and the weights of each edge in the interaction graph, the graph is partitioned with the goal of minimizing the high-dimensional structural entropy after pixel merging in the interaction graph, and the partition result of the interaction graph is obtained; According to the segmentation result of each interaction graph, the superpixel segmentation result of each two-dimensional image is obtained; The superpixel segmentation result of each two-dimensional image is back-projected to obtain the segmentation result of the point cloud data.

[0007] In a possible implementation, based on the distance between pixels in the two-dimensional image, edges are created with the goal of maximizing the one-dimensional structural entropy after the edges are created for the two-dimensional image, to obtain an interaction graph corresponding to the two-dimensional image, including: According to the search radius, an edge is created between each pixel point in the two-dimensional image and the target pixel point to obtain a pending interaction graph, wherein the target pixel point is any pixel point within the search radius range of each pixel point, and the search radius is a preset initial search radius; Calculating the distance between two pixel points corresponding to each edge in the pending interaction graph, and calculating the weight of each edge in the pending interaction graph according to the distance; Calculating the one-dimensional structural entropy of the pending interaction graph obtained within the search radius according to the weight of each edge in the pending interaction graph; Determining whether the increment of the one-dimensional structural entropy is less than a first threshold; If the increment of the one-dimensional structural entropy is greater than or equal to the first threshold, the search radius is increased, and based on the increased search radius, the step of "creating an edge between each pixel point in the two-dimensional image and the target pixel point according to the search radius to obtain a pending interaction graph" and subsequent steps are re-executed; If the increment of the one-dimensional structural entropy is less than the first threshold, the undetermined interaction graph is determined as the interaction graph corresponding to the two-dimensional image.

[0008] In a possible implementation, calculating the distance between two pixel points corresponding to each edge in the pending interaction graph includes: according to , calculate the distance between two pixel points corresponding to each edge in the pending interaction graph; in, is the pixel point in the pending interaction graph and pixels The distance between is the pixel point in the pending interaction graph The color characteristics, is the pixel point in the pending interaction graph The color characteristics, is the pixel point in the pending interaction graph The location characteristics, is the pixel point in the pending interaction graph The location characteristics, and Parameters set according to the image space of the two-dimensional image.

[0009] In a possible implementation, calculating the weight of each edge in the pending interaction graph according to the distance includes: according to , calculate the weight of each edge in the pending interaction graph; in, is the pixel point in the pending interaction graph and pixels The weights of the connected edges, is the normalization term, is the pixel point in the pending interaction graph and pixels The distance between is a parameter that controls the sparsity of the graph.

[0010] In a possible implementation, calculating the one-dimensional structural entropy of the pending interaction graph obtained within the search radius according to the weight of each edge in the pending interaction graph includes: according to , calculating the one-dimensional structural entropy of the undetermined interaction graph obtained under the search radius; in, For the search radius The pending interaction diagram obtained below The one-dimensional structural entropy of The pending interaction diagram The pixels in , The pending interaction diagram The number of pixels in The pending interaction diagram Medium pixel The degree, , The pending interaction diagram Medium pixel and pixels The weights of the connected edges, The pending interaction diagram Center and Pixel The number of all connected edges, For the search radius The pending interaction diagram obtained below The volume of .

[0011] In a possible implementation, the graph is partitioned based on the expected number of superpixels and the weights of the edges in the interaction graph with the goal of minimizing the high-dimensional structural entropy after pixel merging in the interaction graph to obtain the partition result of the interaction graph, including: Initialize each pixel point in the interaction graph as a superpixel partition, obtain the current number of superpixel partitions in the interaction graph, and calculate the current high-dimensional structural entropy of the interaction graph; Determine a current superpixel partition from the superpixel partitions of the interaction graph, and calculate the high-dimensional structural entropy after merging the current superpixel partition with a target superpixel partition directly adjacent to the current superpixel partition according to the weights of the edges in the current superpixel partition and the weights of the edges connecting each pixel point in the current superpixel partition to other superpixel partitions; Calculating the reduction of the high-dimensional structure entropy relative to the current high-dimensional structure entropy; When the reduction is greater than the current maximum reduction, the superpixel partition is merged with the target superpixel partition, the number of current superpixel partitions in the interaction graph is updated, and the high-dimensional structural entropy is used as the new current high-dimensional structural entropy, and the reduction is used as the new current maximum reduction; After the current number of superpixel partitions in the interaction graph is updated, or when the reduction amount is less than or equal to the current maximum reduction amount, determining whether the current number of superpixel partitions reaches the expected number of superpixels; If the number of the current superpixel partitions does not reach the expected number of superpixels, another superpixel partition directly adjacent to the current superpixel partition is used as a new target superpixel partition, or another superpixel partition in the interaction graph is determined as a new current superpixel partition, and the step of "calculating the high-dimensional structural entropy after merging the current superpixel partition with the target superpixel partition directly adjacent to the current superpixel partition according to the weights of each edge in the current superpixel partition and the weights of the edges connecting each pixel point in the current superpixel partition to other superpixel partitions" and subsequent steps are re-executed; If the current number of superpixel partitions reaches the expected number of superpixels, the partition result of the interaction graph is obtained.

[0012] In a possible implementation, the method calculates the high-dimensional structural entropy after merging the current superpixel partition with the target superpixel partition directly adjacent to the current superpixel partition according to the weights of each edge in the current superpixel partition and the weights of the edges connecting each pixel point in the current superpixel partition to other superpixel partitions, including: according to , calculate the high-dimensional structural entropy after merging the current superpixel partition with the target superpixel partition directly adjacent to the current superpixel partition; in, is the high-dimensional structural entropy after merging the current superpixel partition with the target superpixel partition directly adjacent to the current superpixel partition, For this interaction diagram The set of merged superpixel partitions in is the set of superpixel partitions The Superpixel partitions, , is the set of superpixel partitions The number of superpixel partitions in , Partitioning into superpixels The sum of the weights of all edges connecting all pixels in with other superpixel partitions, For interactive diagrams The volume of Partitioning into superpixels The volume of Partitioning into superpixels The The sum of the weights of all edges connecting a pixel to other superpixel partitions.

[0013] In a possible implementation, calculating the reduction of the high-dimensional structure entropy relative to the current high-dimensional structure entropy includes: according to , calculating the reduction of the high-dimensional structure entropy relative to the current high-dimensional structure entropy; in, Partition the current superpixel The target superpixel partition directly adjacent to the current superpixel partition The reduction of the high-dimensional structure entropy after the merger relative to the current high-dimensional structure entropy, is the current high-dimensional structure entropy, Partition the current superpixel The target superpixel partition directly adjacent to the current superpixel partition The high-dimensional structural entropy after merging, For interactive diagrams The volume of Partitioning into superpixels The volume of Partitioning into superpixels The sum of the weights of all edges connecting all pixels in with other superpixel partitions, Partition the target superpixel The volume of Partition the target superpixel The sum of the weights of all edges connecting all pixels in with other superpixel partitions, Partitioning into superpixels Partition with target superpixel The combined volume, Partitioning into superpixels Partition with target superpixel The sum of the weights of all edges connecting all merged pixels to other superpixel partitions.

[0014] In a second aspect, an embodiment of the present invention provides a terminal, comprising a memory and a processor, wherein the memory is used to store a computer program, and the processor is used to call and run the computer program stored in the memory to perform the steps of the method described in the first aspect or any possible implementation of the first aspect.

[0015] In a third aspect, an embodiment of the present invention provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the method described in the first aspect or any possible implementation of the first aspect are implemented.

[0016] The embodiment of the present invention provides a point cloud data segmentation method, terminal and storage medium based on structural information theory, which first projects point cloud data from different dimensions to a two-dimensional plane to obtain each two-dimensional image; then, for each two-dimensional image, based on the distance between pixels in the two-dimensional image, an edge is created with the goal of maximizing the one-dimensional structural entropy after creating the edge for the two-dimensional image, and an interaction graph corresponding to the two-dimensional image is obtained; then, for each interaction graph, based on the expected number of superpixels and the weights of each edge in the interaction graph, the graph is divided with the goal of minimizing the high-dimensional structural entropy after the pixels are merged in the interaction graph, and the division result of the interaction graph is obtained; then, according to the division result of each interaction graph, the superpixel segmentation result of each two-dimensional image is obtained, and the superpixel segmentation result of each two-dimensional image is reversely projected to obtain the segmentation result of the point cloud data. By first projecting the point cloud data from multiple dimensions to a two-dimensional plane, and then introducing the one-dimensional structural entropy and high-dimensional structural entropy in structural information theory, the difficulties of three-dimensional point cloud segmentation in terms of complex topological structure, high computational overhead, and sparse data are overcome, and higher segmentation accuracy, better adaptability and more stable segmentation results can be provided. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0018] Figure 1 is a flow chart of an implementation of a point cloud data segmentation method based on structural information theory provided by an embodiment of the present invention; Figure 2 is a schematic diagram of a process of obtaining an interaction graph provided by an embodiment of the present invention; Figure 3 It is a schematic diagram of a process of dividing an interaction graph provided by an embodiment of the present invention; Figure 4 is a schematic diagram of a process for obtaining a superpixel segmentation result of each two-dimensional image provided by an embodiment of the present invention; Figure 5 is a schematic diagram of the structure of a point cloud data segmentation device based on structural information theory provided by an embodiment of the present invention; Figure 6 is a schematic diagram of a terminal provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0019] In the following description, specific details such as specific system structures, technologies, etc. are provided for the purpose of illustration rather than limitation, so as to provide a thorough understanding of the embodiments of the present invention. However, it should be clear to those skilled in the art that the present invention may be implemented in other embodiments without these specific details. In other cases, detailed descriptions of well-known systems, devices, circuits, and methods are omitted to prevent unnecessary details from obstructing the description of the present invention.

[0020] In order to make the purpose, technical solutions and advantages of the present invention more clear, specific embodiments will be described below in conjunction with the accompanying drawings.

[0021] See also Figure 1 , which shows a flow chart of implementing a point cloud data segmentation method based on structural information theory provided by an embodiment of the present invention, which is described in detail as follows: In step 101, point cloud data is projected from different dimensions onto a two-dimensional plane to obtain respective two-dimensional images.

[0022] Exemplarily, projecting point cloud data from different dimensions onto a two-dimensional plane to obtain respective two-dimensional images may include the following steps: (1) Determine the projection plane and generate a depth map. First, select the xy plane, traverse the point cloud data, project the points onto the xy plane, record the z coordinates to generate a depth map, then select the xz plane, record the y coordinates to generate a depth map, and finally select the yz plane, record the x coordinates to generate a depth map.

[0023] (2) Traverse all projected points, find the maximum and minimum values ​​of the x, y, and z coordinates on each plane, and determine the width and height of the two-dimensional plane. Taking the xy plane as an example, the width of the two-dimensional plane can be obtained based on the difference between the maximum and minimum values ​​of the x coordinate, and the height of the two-dimensional plane can be obtained based on the difference between the maximum and minimum values ​​of the y coordinate.

[0024] (3) Determine the actual length L represented by a single pixel in the width and height directions based on the projection range, the expected image range, and the image resolution. Calculate the actual length of each pixel in the width direction, according to Calculate the actual length that each pixel represents in the height direction, where the image width and image height are the width and height of the expected image range.

[0025] (4) Traverse the point cloud and calculate the coordinates of each pixel in the two-dimensional image.

[0026] Among them, the coordinates of each pixel in the two-dimensional image for: ; ; in, is the minimum value in the width direction of the two-dimensional image, It is the minimum value in the height direction of the two-dimensional image.

[0027] Then, the color information contained in the point cloud data is associated with the coordinates of each pixel in the corresponding two-dimensional image to achieve color information filling, and the image is saved in the required image format to obtain a two-dimensional image.

[0028] In this embodiment, processing the point cloud data into two-dimensional images for segmentation can simplify data processing, reduce computational complexity, improve segmentation efficiency and accuracy, and support rich information such as depth maps. By mapping multiple views and combining the segmentation results, the segmentation quality and accuracy of the point cloud data can be significantly improved, and the complete three-dimensional structure can be captured, information loss can be reduced, segmentation accuracy can be improved, sparse area processing capabilities can be enhanced, and the robustness of the algorithm and the ability to process complex geometric shapes can be improved. This multi-view projection method can not only help achieve more refined point cloud segmentation, but also improve the accuracy and consistency of point cloud segmentation in three-dimensional reconstruction.

[0029] In step 102, for each two-dimensional image, based on the distance between pixels in the two-dimensional image, an edge is created with the goal of maximizing the one-dimensional structural entropy after creating the edge for the two-dimensional image, to obtain an interaction graph corresponding to the two-dimensional image.

[0030] Optionally, step 102 may include: According to the search radius, an edge is created between each pixel point in the two-dimensional image and the target pixel point to obtain a pending interaction graph, wherein the target pixel point is any pixel point within the search radius of each pixel point, and the search radius is a preset initial search radius.

[0031] The distance between two pixel points corresponding to each edge in the pending interaction graph is calculated, and the weight of each edge in the pending interaction graph is calculated based on the distance.

[0032] According to the weight of each edge in the pending interaction graph, the one-dimensional structural entropy of the pending interaction graph obtained within the search radius is calculated.

[0033] It is determined whether the increment of the one-dimensional structural entropy is less than a first threshold.

[0034] If the increment of the one-dimensional structural entropy is greater than or equal to the first threshold, the search radius is increased, and based on the increased search radius, the step of "creating an edge between each pixel point in the two-dimensional image and the target pixel point according to the search radius to obtain a pending interaction graph" and subsequent steps are re-executed.

[0035] If the increment of the one-dimensional structural entropy is less than the first threshold, the to-be-determined interaction graph is determined as the interaction graph corresponding to the two-dimensional image.

[0036] Exemplarily, calculating the distance between two pixel points corresponding to each edge in the pending interaction graph may include: according to , calculate the distance between the two pixel points corresponding to each edge in the pending interaction graph.

[0037] in, The pixel point in the pending interaction map and pixels The distance between The pixel point in the pending interaction map The color characteristics, The pixel point in the pending interaction map The color characteristics, The pixel point in the pending interaction map The location characteristics, The pixel point in the pending interaction map The location characteristics, and Parameters set according to the image space of the two-dimensional image.

[0038] Exemplarily, calculating the weight of each edge in the pending interaction graph according to the distance may include: according to , calculate the weight of each edge in the pending interaction graph.

[0039] in, The pixel point in the pending interaction map and pixels The weights of the connected edges, is the normalization term, The pixel point in the pending interaction map and pixels The distance between is a parameter that controls the sparsity of the graph.

[0040] Exemplarily, according to the weight of each edge in the pending interaction graph, calculating the one-dimensional structural entropy of the pending interaction graph obtained within the search radius may include: according to , calculate the one-dimensional structural entropy of the undetermined interaction graph obtained under the search radius.

[0041] in, For the search radius The pending interaction diagram obtained below The one-dimensional structural entropy of To be determined interaction diagram The pixels in , To be determined interaction diagram The number of pixels in To be determined interaction diagram Medium pixel The degree, , To be determined interaction diagram Medium pixel and pixels The weights of the connected edges, To be determined interaction diagram Center and Pixel The number of all connected edges, For the search radius The pending interaction diagram obtained below The volume of .

[0042] Exemplary, combined Figure 2 , assuming that the input two-dimensional image is , the process of obtaining the interaction graph corresponding to the two-dimensional image can be: Input: 2D image , a parameter used to adjust the sensitivity of pixel similarity to distance calculation , the judgment standard threshold of one-dimensional structure entropy change (also known as the first threshold) .

[0043] Initial 2D image There are no edges between the pixels in the graph, which is a borderless graph. By gradually increasing the search radius To explore the pixel neighborhood, Start, increase in sequence, for each pixel , traverse its search radius Neighborhood below Each pixel within Among them, the neighborhood Defined as the distance pixel for The set of pixels within the range, With Neighborhood Each pixel in is connected. For example, when When the pixel It is connected to the adjacent upper, lower, left, right, upper left, upper right, lower left, and lower right pixels. Increase the neighborhood Expand outward in a similar manner to include pixels that are further away.

[0044] When determining the search radius And each pixel With Neighborhood After each pixel point in the The distance between them is defined by the color feature distance and the position feature distance, which measures the difference between the pixel pairs in color and spatial position.

[0045] Calculate the search radius Each pixel pair connected below After the distance between them is calculated, each connected pixel pair is calculated based on the distance The weight of the edge formed.

[0046] Then based on the search radius The weight of each edge is calculated according to Calculate the one-dimensional structural entropy when , which means that as the search radius increases, the one-dimensional structure entropy grows slowly. Continuing to increase the search radius will have limited effect on improving the graph information. Therefore, we should stop increasing the search radius and terminate the search radius. The loop completes the graph construction phase and converts the input two-dimensional image Mapping to Graph , and get the interaction diagram ,in, Corresponding to each pixel in the interaction graph, is the set of interconnected edges in the interaction graph, is the weight matrix of the interconnected edges in the interaction graph, which is used to measure the similarity between pixels.

[0047] In this embodiment, traditional graph generation methods usually connect nodes according to fixed rules, such as distance-based thresholds or K-nearest neighbor methods. These methods are prone to excessive information loss (over-simplification of the graph) or generation of redundant structures (over-complexity of the graph) in terms of information retention. They are unable to adjust the connection method according to local complexity, resulting in the same strategy being used for both complex and simple areas. This can easily lead to loss of details in complex areas or generation of redundant structures in simple areas. The size of the neighborhood usually needs to be manually specified and cannot adapt to complex changes in data.

[0048] This embodiment uses the increment of one-dimensional structural entropy to make judgments, maximize the one-dimensional structural entropy to construct an interaction graph, and incremental construction means that the structural entropy is gradually maximized. The system gradually optimizes the structural information of the graph with low computational overhead. The incremental one-dimensional structural entropy gradually captures the interaction between distant pixels by expanding the neighborhood of the nodes layer by layer, enhancing the retention of global information, so that the graph can better reflect the long-distance relationship in the image. By incrementally adding edges, the system can control the complexity of the graph, avoid generating an overly dense graph structure, reduce the number of edges in the graph while retaining sufficient information, and make the generated graph more concise and efficient.

[0049] In step 103, for each interaction graph, based on the expected number of superpixels and the weights of each edge in the interaction graph, the graph is partitioned with the goal of minimizing the high-dimensional structural entropy after pixel merging in the interaction graph to obtain a partition result of the interaction graph.

[0050] Optionally, step 103 may include: Each pixel point in the interaction graph is initialized as a superpixel partition, the current number of superpixel partitions in the interaction graph is obtained, and the current high-dimensional structural entropy of the interaction graph is calculated.

[0051] A current superpixel partition is determined from the superpixel partitions of the interaction graph, and the high-dimensional structural entropy of the current superpixel partition and the target superpixel partition directly adjacent to the current superpixel partition are calculated according to the weights of each edge in the current superpixel partition and the weights of the edges connecting each pixel in the current superpixel partition to other superpixel partitions.

[0052] Calculate the reduction of the high-dimensional structure entropy relative to the current high-dimensional structure entropy.

[0053] When the reduction is greater than the current maximum reduction, the superpixel partition is merged with the target superpixel partition, the current number of superpixel partitions in the interaction graph is updated, and the high-dimensional structural entropy is used as the new current high-dimensional structural entropy, and the reduction is used as the new current maximum reduction.

[0054] After the current number of superpixel partitions in the interaction graph is updated, or when the reduction amount is less than or equal to the current maximum reduction amount, it is determined whether the current number of superpixel partitions reaches the expected number of superpixels.

[0055] If the number of current superpixel partitions does not reach the expected number of superpixels, another superpixel partition directly adjacent to the current superpixel partition is used as a new target superpixel partition, or another superpixel partition in the interaction graph is determined as a new current superpixel partition, and the step of "calculating the high-dimensional structural entropy after merging the current superpixel partition with the target superpixel partition directly adjacent to the current superpixel partition based on the weights of each edge in the current superpixel partition and the weights of the edges connecting each pixel point in the current superpixel partition to other superpixel partitions" and subsequent steps are re-executed.

[0056] If the current number of superpixel partitions reaches the expected number of superpixels, the partition result of the interaction graph is obtained.

[0057] For example, according to , calculate the high-dimensional structural entropy after merging the current superpixel partition with the target superpixel partition directly adjacent to the current superpixel partition.

[0058] in, is the high-dimensional structural entropy after merging the current superpixel partition with the target superpixel partition directly adjacent to the current superpixel partition, For this interaction diagram The set of merged superpixel partitions in is the set of superpixel partitions The Superpixel partitions, , is the set of superpixel partitions The number of superpixel partitions in , Partitioning into superpixels The sum of the weights of all edges connecting all pixels in with other superpixel partitions, For interactive diagrams The volume can be referred to calculate, Partitioning into superpixels The volume can also be referred to calculate, Partitioning into superpixels The The sum of the weights of all edges connecting a pixel to other superpixel partitions.

[0059] Exemplarily, calculating the reduction of the high-dimensional structure entropy relative to the current high-dimensional structure entropy may include: according to , calculate the reduction of the high-dimensional structure entropy relative to the current high-dimensional structure entropy.

[0060] in, Partition the current superpixel The target superpixel partition directly adjacent to the current superpixel partition The reduction of the high-dimensional structure entropy after the merger relative to the current high-dimensional structure entropy, is the current high-dimensional structure entropy, Partition the current superpixel The target superpixel partition directly adjacent to the current superpixel partition The high-dimensional structural entropy after merging, For interactive diagrams The volume of Partitioning into superpixels The volume of Partitioning into superpixels The sum of the weights of all edges connecting all pixels in with other superpixel partitions, Partition the target superpixel The volume of Partition the target superpixel The sum of the weights of all edges connecting all pixels in with other superpixel partitions, Partitioning into superpixels Partition with target superpixel The combined volume, Partitioning into superpixels Partition with target superpixel The sum of the weights of all edges connecting all merged pixels to other superpixel partitions.

[0061] Exemplary, combined Figure 3 , for the interaction graph The process of graph partitioning can be: Initialize superpixel partitioning and enter the expected number of superpixel targets (that is, the expected number of superpixels) , the interaction diagram Each pixel in the image is initialized as an independent superpixel partition, that is, each pixel in the image is initially a superpixel partition, which provides an initial state for subsequent merging operations and builds a superpixel hierarchy starting from the finest pixel level.

[0062] Iterative merging loop, start a loop for iterative merging of superpixels until a specific termination condition is met (i.e. the number of current superpixel partitions reaches the expected number of superpixels) ): Traverse the superpixel partitions to find the merge target, and traverse each superpixel partition in all current superpixel partitions , for each superpixel partition , initialize its target merged superpixel partition Empty, and the current maximum reduction 0, traverse the superpixel partition All directly adjacent superpixels of ; Calculate superpixel partitions Partitioning with neighboring superpixels The reduction of high-dimensional structural entropy after merging If the reduction is greater than the current maximum reduction recorded , update the current maximum reduction to the current , and update the target merged superpixel partition For the current .

[0063] Merge and update to partition superpixels Merge superpixel partitions with their targets Merge to form a new superpixel partition ,This is a key step in constructing a hierarchical superpixel structure, which advances the superpixel partitioning hierarchy from fine-grained to coarse-grained by continuously merging adjacent superpixel partitions.

[0064] Among them, for a pair of directly adjacent superpixels and ,merge and Interaction diagram Adding a new superpixel , whose child nodes are and The union of the child nodes, then from the interaction graph Remove and , after merging, the new partition becomes ,in .

[0065] Check the termination condition and return the result, check whether the number of superpixels in the current superpixel partition set reaches the expected number of superpixels , if the desired number of superpixels is reached , the final superpixel partition set is returned to complete the superpixel segmentation process and obtain the hierarchical superpixel partition result that meets the requirements; if it is not reached, the merging operation is continued in an iterative loop.

[0066] Among them, in the interaction diagram When each pixel in is initialized as an independent superpixel partition and the pixels are merged, the corresponding two-dimensional structural entropy is ,at this time If the interaction diagram After merging each pixel in the image, the expected number of superpixels is still not reached. , then continue to merge at the next level, which corresponds to the three-dimensional structure entropy, and so on.

[0067] Among them, taking the two-dimensional structure entropy as an example, the reduction of the two-dimensional structure entropy after merging is selected Greater than the current maximum reduction recorded The target superpixel partition is selected as follows: .

[0068] in, Partitioning into superpixels The goal is to merge superpixel partitions, Partitioning into superpixels The set of directly adjacent superpixels of .

[0069] Accordingly, the superpixel is partitioned Merge superpixel partitions with the target Merge into new superpixel partitions Add to Interaction Diagram Its child nodes are and The union of the child nodes and from the interaction graph Remove and .

[0070] This embodiment considers traditional graph partitioning methods, such as the partitioning method based on the minimum cut set, K-means clustering or spectral clustering, which often only consider local information in the graph or are based on a certain fixed optimization criterion, only focus on local features, and ignore the global information of the graph. This locality may lead to the inability to accurately capture global features or long-range dependencies in complex networks, and it is easy to lose important structural information or cause the partitioning results to be overly simplified. In addition, the traditional graph partitioning method requires the number of clusters or the number of groups to be divided to be specified in advance. This method that relies on user settings may lead to unreasonable partitioning, especially when the structure of the graph is not clear or there is no clear number of clusters. This mandatory setting will lead to inaccurate partitioning and cannot handle the multi-level structure in the graph well. Especially when it is necessary to process information of multiple scales at the same time, the fixed hierarchical structure partitioning method is prone to errors.

[0071] The graph partitioning method based on high-dimensional structural entropy not only focuses on the similarity of local nodes, but also makes full use of the global information of the graph. Through the optimization of high-dimensional structural entropy, the system can capture the mutual relationship between remote nodes, not just the local relationship between adjacent nodes, which makes the graph partitioning result better reflect the overall structure of the data. By minimizing the high-dimensional structural entropy, graph partitioning can better preserve the information structure of the original graph, while ensuring that the sub-graphs after partitioning can capture important patterns and laws. Hierarchical partitioning can adapt to features of different scales, which is particularly suitable for multi-level tasks such as image segmentation, reducing information loss in the partitioning process and better retaining key information in the original data.

[0072] In step 104, a superpixel segmentation result of each two-dimensional image is obtained according to the segmentation result of each interaction graph.

[0073] This embodiment takes into account that traditional single segmentation methods usually can only work at a certain scale, and it is difficult to process the global structure and detail features of the image at the same time. This leads to the segmentation results being too rough or losing key details when processing complex scenes. Segmentation can only be performed based on local information or global information, and it is difficult to take both into account at the same time. This leads to some methods losing the global structure when capturing details, or sacrificing local details while retaining the global structure.

[0074] In this embodiment, combined with Figure 4,In the graph partition stage, the optimal coding tree is generated by minimizing the structural entropy (i.e. maximizing the decoding information).,Each node in the optimal coding tree corresponds to a region of the image.,These regions represent segmentation results of different scales at different levels.,The segmentation regions at different levels are integrated to reconstruct the segmentation of the original image.,During the reconstruction, it is ensured that the pixels inside each superpixel are connected to form a continuous region that is not,interrupted by the pixels of other superpixels.

[0075] In step 105, the superpixel segmentation result of each two-dimensional image is back-projected to obtain the segmentation result of the point cloud data.

[0076] Exemplarily, the process of back-projection may be: Traverse each pixel of the two-dimensional image to obtain color information and depth value. That is, for each perspective, for each pixel , get its color information and depth value (that is, z value, y value or x value), and then record it.

[0077] For each 2D image of each perspective, its corresponding pixel coordinates and depth values ​​are back-projected to the real coordinates of the 3D point cloud, where: xy viewing angle: corresponding image coordinates Map to coordinate, The coordinates are recovered from the depth map: .

[0078] xz viewing angle: .

[0079] YZ perspective: .

[0080] in, , , They are the pixels of the two-dimensional image at xy, xz and yz perspectives respectively. Corresponding stored depth value.

[0081] Restore point cloud data and process overlapping points. When the point cloud results of inverse mapping from different perspectives overlap, process the conflict in the overlapping area. Among the results from different perspectives, select the closest depth value, retain the closest point, and discard the points far away from the observer.

[0082] Traverse the pixels of all viewpoints, inversely map them back to 3D point clouds, and finally merge the 3D point data to achieve complete point cloud segmentation.

[0083] This embodiment maps the point cloud data to a two-dimensional plane for processing, fully considering the problem of large amount of computation and high complexity of directly segmenting the point cloud data. The superpixel segmentation based on structural information theory retains the advantages of ordinary superpixel segmentation, and considers adding structural information theory in the graph generation and graph division stages, making full use of the global information of the graph, being able to adaptively determine the optimal number of divisions, processing complex multi-level structures, and realizing divisions of different scales. It can provide a deeper and more accurate explanation of segmentation, which helps to better process and solve point cloud data segmentation tasks.

[0084] The point cloud data segmentation method based on structural information theory provided by the embodiment of the present invention maps multiple views of point cloud data onto a two-dimensional plane, overcoming the difficulties of three-dimensional point cloud segmentation in terms of complex topological structure, high computational overhead, and sparse data. At the same time, it also simplifies the processing process, reduces the amount of data to be processed, improves computational efficiency, and can better fit the edges and boundaries of objects in the image, making it easier to combine other modal data to achieve efficient and accurate segmentation. In addition, on the basis of retaining the advantages of traditional superpixel segmentation, considering adding structural information theory, by maximizing the decoded information (that is, minimizing the high-dimensional structural entropy) to obtain the optimal graph partitioning result, it can reduce information loss during the segmentation process, and through hierarchical superpixel segmentation, it can adapt to different application requirements, thereby effectively utilizing global information, avoiding overly complex graph structures while maintaining graph information, and realizing multi-scale segmentation, it can also maximize information retention, improve boundary fit, and adapt to a variety of application scenarios, thereby expanding the favorable characteristics of superpixel segmentation and solving problems such as high data processing complexity, unstable and inaccurate segmentation results. . Compared with the traditional superpixel segmentation method, it provides higher segmentation accuracy, better adaptability and more stable segmentation results, and has further improved the performance, making it more suitable for segmenting point cloud data.

[0085] It should be understood that the order of execution of the steps in the above embodiment does not necessarily mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiment of the present invention.

[0086] The following is an embodiment of the device of the present invention. For details not described in detail therein, reference may be made to the corresponding method embodiment described above.

[0087] Figure 5 The structure diagram of the point cloud data segmentation device based on structural information theory provided by an embodiment of the present invention is shown. For the convenience of explanation, only the part related to the embodiment of the present invention is shown, which is described in detail as follows: like Figure 5As shown, the point cloud data segmentation device based on structural information theory includes: a first processing module 51, a second processing module 52, a third processing module 53, a fourth processing module 54 and a fifth processing module 55.

[0088] The first processing module 51 is used to project the point cloud data from different dimensions onto a two-dimensional plane to obtain respective two-dimensional images.

[0089] The second processing module 52 is used to create edges for each two-dimensional image based on the distance between pixels in the two-dimensional image with the goal of maximizing the one-dimensional structural entropy after creating the edges for the two-dimensional image, so as to obtain an interaction graph corresponding to the two-dimensional image.

[0090] The third processing module 53 is used to perform graph partitioning for each interaction graph based on the expected number of superpixels and the weights of each edge in the interaction graph, with the goal of minimizing the high-dimensional structural entropy after pixel merging in the interaction graph, to obtain a partitioning result of the interaction graph.

[0091] The fourth processing module 54 is used to obtain the superpixel segmentation result of each two-dimensional image according to the segmentation result of each interaction graph.

[0092] The fifth processing module 55 is used to perform back-projection on the superpixel segmentation result of each two-dimensional image to obtain the segmentation result of the point cloud data.

[0093] The embodiment of the present invention overcomes the difficulties of three-dimensional point cloud segmentation in terms of complex topological structure, high computational overhead, and sparse data by mapping multiple views of point cloud data onto a two-dimensional plane. At the same time, it also simplifies the processing process, reduces the amount of data to be processed, improves computational efficiency, and can better fit the edges and boundaries of objects in the image, making it easier to combine other modal data to achieve efficient and accurate segmentation. In addition, on the basis of retaining the advantages of traditional superpixel segmentation, structural information theory is considered to be added to obtain the optimal graph partitioning result by maximizing the decoded information (that is, minimizing the high-dimensional structural entropy), which can reduce information loss during the segmentation process. Through hierarchical superpixel segmentation, it can adapt to different application requirements, thereby effectively utilizing global information, avoiding overly complex graph structures while maintaining graph information, and realizing multi-scale segmentation. It can also maximize information retention, improve boundary fit, and adapt to a variety of application scenarios, thereby expanding the favorable characteristics of superpixel segmentation and solving problems such as high data processing complexity, unstable and inaccurate segmentation results. . Compared with the traditional superpixel segmentation method, it provides higher segmentation accuracy, better adaptability and more stable segmentation results, and has further improved the performance, making it more suitable for segmenting point cloud data.

[0094] In a possible implementation, the second processing module 52 can be used to create an edge between each pixel point in the two-dimensional image and a target pixel point according to a search radius to obtain a pending interaction graph, wherein the target pixel point is any pixel point within the search radius of each pixel point, and the search radius is a preset initial search radius; calculate the distance between two pixels corresponding to each edge in the pending interaction graph, and calculate the weight of each edge in the pending interaction graph according to the distance; calculate the one-dimensional structural entropy of the pending interaction graph obtained under the search radius according to the weight of each edge in the pending interaction graph; determine whether the increment of the one-dimensional structural entropy is less than a first threshold; if the increment of the one-dimensional structural entropy is greater than or equal to the first threshold, increase the search radius, and re-execute the step of "creating an edge between each pixel point in the two-dimensional image and the target pixel point according to the search radius to obtain a pending interaction graph" and subsequent steps based on the increased search radius; if the increment of the one-dimensional structural entropy is less than the first threshold, determine the pending interaction graph as the interaction graph corresponding to the two-dimensional image.

[0095] In a possible implementation, the second processing module 52 may be configured to: , calculate the distance between two pixel points corresponding to each edge in the pending interaction graph; in, is the pixel point in the pending interaction graph and pixels The distance between is the pixel point in the pending interaction graph The color characteristics, is the pixel point in the pending interaction graph The color characteristics, is the pixel point in the pending interaction graph The location characteristics, is the pixel point in the pending interaction graph The location characteristics, and Parameters set according to the image space of the two-dimensional image.

[0096] In a possible implementation, the second processing module 52 may be configured to: , calculate the weight of each edge in the pending interaction graph; in, is the pixel point in the pending interaction graph and pixels The weights of the connected edges, is the normalization term, is the pixel point in the pending interaction graph and pixels The distance between is a parameter that controls the sparsity of the graph.

[0097] In a possible implementation, the second processing module 52 may be configured to: , calculating the one-dimensional structural entropy of the undetermined interaction graph obtained under the search radius; in, For the search radius The pending interaction diagram obtained below The one-dimensional structural entropy of The pending interaction diagram The pixels in , The pending interaction diagram The number of pixels in The pending interaction diagram Medium pixel The degree, , The pending interaction diagram Medium pixel and pixels The weights of the connected edges, The pending interaction diagram Center and Pixel The number of all connected edges, For the search radius The pending interaction diagram obtained below The volume of .

[0098] In one possible implementation, the third processing module 53 can be used to determine a current superpixel partition from the superpixel partitions of the interaction graph, and calculate the high-dimensional structural entropy after merging the current superpixel partition with the target superpixel partition directly adjacent to the current superpixel partition based on the weights of the edges in the current superpixel partition and the weights of the edges connecting each pixel in the current superpixel partition to other superpixel partitions; calculate the reduction in the high-dimensional structural entropy relative to the current high-dimensional structural entropy; when the reduction is greater than the current maximum reduction, merge the superpixel partition with the target superpixel partition, update the number of current superpixel partitions in the interaction graph, and use the high-dimensional structural entropy as the new current high-dimensional structural entropy, and use the reduction as the new current maximum reduction; when updating the current superpixel partition in the interaction graph After the number of superpixel partitions is reached, or when the reduction is less than or equal to the current maximum reduction, determine whether the current number of superpixel partitions reaches the expected number of superpixels; if the current number of superpixel partitions does not reach the expected number of superpixels, take another superpixel partition directly adjacent to the current superpixel partition as the new target superpixel partition, or determine another superpixel partition in the interaction graph as the new current superpixel partition, and re-execute the step of "calculating the high-dimensional structural entropy after merging the current superpixel partition with the target superpixel partition directly adjacent to the current superpixel partition according to the weights of each edge in the current superpixel partition and the weights of the edges connecting each pixel point in the current superpixel partition with other superpixel partitions" and subsequent steps; if the current number of superpixel partitions reaches the expected number of superpixels, obtain the division result of the interaction graph.

[0099] In a possible implementation, the third processing module 53 may be configured to: , calculate the high-dimensional structural entropy after merging the current superpixel partition with the target superpixel partition directly adjacent to the current superpixel partition; in, is the high-dimensional structural entropy after merging the current superpixel partition with the target superpixel partition directly adjacent to the current superpixel partition, For this interaction diagram The set of merged superpixel partitions in is the set of superpixel partitions The Superpixel partitions, , is the set of superpixel partitions The number of superpixel partitions in , Partitioning into superpixels The sum of the weights of all edges connecting all pixels in with other superpixel partitions, For interactive diagrams The volume of Partitioning into superpixels The volume of Partitioning into superpixels The The sum of the weights of all edges connecting a pixel to other superpixel partitions.

[0100] In a possible implementation, the third processing module 53 may be configured to: , calculating the reduction of the high-dimensional structure entropy relative to the current high-dimensional structure entropy; in, Partition the current superpixel The target superpixel partition directly adjacent to the current superpixel partition The reduction of the high-dimensional structure entropy after the merger relative to the current high-dimensional structure entropy, is the current high-dimensional structure entropy, Partition the current superpixel The target superpixel partition directly adjacent to the current superpixel partition The high-dimensional structural entropy after merging, For interactive diagrams The volume of Partitioning into superpixels The volume of Partitioning into superpixels The sum of the weights of all edges connecting all pixels in with other superpixel partitions, Partition the target superpixel The volume of Partition the target superpixel The sum of the weights of all edges connecting all pixels in with other superpixel partitions, Partitioning into superpixels Partition with target superpixel The combined volume, Partitioning into superpixels Partition with target superpixel The sum of the weights of all edges connecting all merged pixels to other superpixel partitions.

[0101] Figure 6 is a schematic diagram of a terminal provided by an embodiment of the present invention. Figure 6 As shown, the terminal 6 of this embodiment includes: a processor 60, a memory 61, and a computer program 62 stored in the memory 61 and executable on the processor 60. When the processor 60 executes the computer program 62, the steps in the above-mentioned point cloud data segmentation method embodiments based on structural information theory are implemented, for example Figure 1 Alternatively, when the processor 60 executes the computer program 62, the functions of each module / unit in the above-mentioned device embodiments are implemented, for example Figure 5Functions of the modules / units 51 to 55 are shown.

[0102] Exemplarily, the computer program 62 may be divided into one or more modules / units, one or more modules / units are stored in the memory 61 and executed by the processor 60 to implement the present invention. One or more modules / units may be a series of computer program instruction segments capable of implementing specific functions, and the instruction segments are used to describe the execution process of the computer program 62 in the terminal 6. For example, the computer program 62 may be divided into Figure 5 Modules / units 51 to 55 are shown.

[0103] The terminal 6 may be a computing device such as a desktop computer, a notebook, a PDA, or a cloud server. The terminal 6 may include, but is not limited to, a processor 60 and a memory 61. Those skilled in the art will appreciate that Figure 6 It is only an example of terminal 6 and does not constitute a limitation on terminal 6. It may include more or fewer components than shown in the figure, or a combination of certain components, or different components. For example, the terminal may also include input and output devices, network access devices, buses, etc.

[0104] The processor 60 may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor may be a microprocessor or any conventional processor, etc.

[0105] The memory 61 may be an internal storage unit of the terminal 6, such as a hard disk or memory of the terminal 6. The memory 61 may also be an external storage device of the terminal 6, such as a plug-in hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card, etc. equipped on the terminal 6. Further, the memory 61 may also include both an internal storage unit of the terminal 6 and an external storage device. The memory 61 is used to store computer programs and other programs and data required by the terminal. The memory 61 may also be used to temporarily store data that has been output or is to be output.

[0106] The technicians in the relevant field can clearly understand that for the convenience and simplicity of description, only the division of the above-mentioned functional units and modules is used as an example for illustration. In practical applications, the above-mentioned function allocation can be completed by different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above. The functional units and modules in the embodiment can be integrated in a processing unit, or each unit can exist physically separately, or two or more units can be integrated in one unit. The above-mentioned integrated unit can be implemented in the form of hardware or in the form of software functional units. In addition, the specific names of the functional units and modules are only for the convenience of distinguishing each other, and are not used to limit the scope of protection of this application. The specific working process of the units and modules in the above-mentioned system can refer to the corresponding process in the aforementioned method embodiment, which will not be repeated here.

[0107] In the above embodiments, the description of each embodiment has its own emphasis. For parts that are not described or recorded in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0108] Those skilled in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of the present invention.

[0109] In the embodiments provided by the present invention, it should be understood that the disclosed devices / terminals and methods can be implemented in other ways. For example, the device / terminal embodiments described above are only schematic, for example, the division of modules or units is only a logical function division, and there may be other division methods in actual implementation, such as 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 mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0110] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0111] In addition, each functional unit in each embodiment of the present invention may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0112] If the integrated module / 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 this understanding, the present invention implements all or part of the processes in the above-mentioned embodiment method, and can also be completed by instructing the relevant hardware through a computer program. The computer program can be stored in a computer-readable storage medium. When the computer program is executed by the processor, it can implement the steps of the above-mentioned point cloud data segmentation method embodiments based on structural information theory. Among them, the computer program includes computer program code, and the computer program code can be in source code form, object code form, executable file or some intermediate form. Computer-readable media can include: any entity or device that can carry computer program code, recording medium, U disk, mobile hard disk, disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signal, telecommunication signal and software distribution medium.

[0113] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features may be replaced by equivalents. Such modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention, and should be included in the protection scope of the present invention.

Claims

1. A point cloud data segmentation method based on structural information theory, characterized in that: include: Project the point cloud data from different dimensions onto a two-dimensional plane to obtain individual two-dimensional images; For each two-dimensional image, based on the distance between pixels in the two-dimensional image, an edge is created with the goal of maximizing the one-dimensional structural entropy after the edge is created for the two-dimensional image, and an interaction graph corresponding to the two-dimensional image is obtained; For each interaction graph, based on the expected number of superpixels and the weights of each edge in the interaction graph, the graph is partitioned with the goal of minimizing the high-dimensional structural entropy after pixel merging in the interaction graph, and the partition result of the interaction graph is obtained; According to the segmentation result of each interaction graph, the superpixel segmentation result of each two-dimensional image is obtained; The superpixel segmentation result of each two-dimensional image is back-projected to obtain the segmentation result of the point cloud data.

2. The point cloud data segmentation method based on structural information theory according to claim 1, characterized in that: Based on the distance between pixels in the two-dimensional image, edges are created with the goal of maximizing the one-dimensional structural entropy after the edges are created for the two-dimensional image, and an interaction graph corresponding to the two-dimensional image is obtained, including: According to the search radius, an edge is created between each pixel point in the two-dimensional image and the target pixel point to obtain a pending interaction graph, wherein the target pixel point is any pixel point within the search radius range of each pixel point, and the search radius is a preset initial search radius; Calculating the distance between two pixel points corresponding to each edge in the pending interaction graph, and calculating the weight of each edge in the pending interaction graph according to the distance; Calculating the one-dimensional structural entropy of the pending interaction graph obtained within the search radius according to the weight of each edge in the pending interaction graph; Determining whether the increment of the one-dimensional structural entropy is less than a first threshold; If the increment of the one-dimensional structural entropy is greater than or equal to the first threshold, the search radius is incremented, and based on the incremented search radius, the step of "creating an edge between each pixel point in the two-dimensional image and the target pixel point according to the search radius to obtain a pending interaction graph" and subsequent steps are re-executed; If the increment of the one-dimensional structural entropy is less than the first threshold, the undetermined interaction graph is determined as the interaction graph corresponding to the two-dimensional image.

3. The point cloud data segmentation method based on structural information theory according to claim 2, characterized in that: Calculating the distance between two pixel points corresponding to each edge in the pending interaction graph includes: according to , calculate the distance between two pixel points corresponding to each edge in the pending interaction graph; in, is the pixel point in the pending interaction graph and pixels The distance between is the pixel point in the pending interaction graph The color characteristics, is the pixel point in the pending interaction graph The color characteristics, is the pixel point in the pending interaction graph The location characteristics, is the pixel point in the pending interaction graph The location characteristics, and Parameters set according to the image space of the two-dimensional image.

4. The point cloud data segmentation method based on structural information theory according to claim 2, characterized in that: Calculating the weight of each edge in the pending interaction graph according to the distance includes: according to , calculate the weight of each edge in the pending interaction graph; in, is the pixel point in the pending interaction graph and pixels The weights of the connected edges, is the normalization term, is the pixel point in the pending interaction graph and pixels The distance between is a parameter that controls the sparsity of the graph.

5. The point cloud data segmentation method based on structural information theory according to claim 2, characterized in that: Calculating the one-dimensional structural entropy of the pending interaction graph obtained within the search radius according to the weight of each edge in the pending interaction graph includes: according to , calculating the one-dimensional structural entropy of the undetermined interaction graph obtained under the search radius; in, For the search radius The pending interaction diagram obtained below The one-dimensional structural entropy of The pending interaction diagram The pixels in , The pending interaction diagram The number of pixels in The pending interaction diagram Medium pixel The degree, , The pending interaction diagram Medium pixel and pixels The weights of the connected edges, The pending interaction diagram Center and Pixel The number of all connected edges, For the search radius The pending interaction diagram obtained below The volume of .

6. The point cloud data segmentation method based on structural information theory according to claim 1, characterized in that: The method of performing graph partitioning based on the expected number of superpixels and the weights of the edges in the interaction graph with the goal of minimizing the high-dimensional structural entropy after pixel merging in the interaction graph to obtain the partitioning result of the interaction graph includes: Initialize each pixel point in the interaction graph as a superpixel partition, obtain the current number of superpixel partitions in the interaction graph, and calculate the current high-dimensional structural entropy of the interaction graph; Determine a current superpixel partition from the superpixel partitions of the interaction graph, and calculate the high-dimensional structural entropy after merging the current superpixel partition with a target superpixel partition directly adjacent to the current superpixel partition according to the weights of the edges in the current superpixel partition and the weights of the edges connecting each pixel point in the current superpixel partition to other superpixel partitions; Calculating the reduction of the high-dimensional structure entropy relative to the current high-dimensional structure entropy; When the reduction is greater than the current maximum reduction, the superpixel partition is merged with the target superpixel partition, the number of current superpixel partitions in the interaction graph is updated, and the high-dimensional structural entropy is used as the new current high-dimensional structural entropy, and the reduction is used as the new current maximum reduction; After the current number of superpixel partitions in the interaction graph is updated, or when the reduction amount is less than or equal to the current maximum reduction amount, determining whether the current number of superpixel partitions reaches the expected number of superpixels; If the number of the current superpixel partitions does not reach the expected number of superpixels, another superpixel partition directly adjacent to the current superpixel partition is used as a new target superpixel partition, or another superpixel partition in the interaction graph is determined as a new current superpixel partition, and the step of "calculating the high-dimensional structural entropy after merging the current superpixel partition with the target superpixel partition directly adjacent to the current superpixel partition according to the weights of each edge in the current superpixel partition and the weights of the edges connecting each pixel point in the current superpixel partition to other superpixel partitions" and subsequent steps are re-executed; If the current number of superpixel partitions reaches the expected number of superpixels, the partition result of the interaction graph is obtained.

7. The point cloud data segmentation method based on structural information theory according to claim 6, characterized in that: The method calculates the high-dimensional structural entropy after merging the current superpixel partition with the target superpixel partition directly adjacent to the current superpixel partition according to the weights of each edge in the current superpixel partition and the weights of the edges connecting each pixel point in the current superpixel partition to other superpixel partitions, including: according to , calculate the high-dimensional structural entropy after merging the current superpixel partition with the target superpixel partition directly adjacent to the current superpixel partition; in, is the high-dimensional structural entropy after merging the current superpixel partition with the target superpixel partition directly adjacent to the current superpixel partition, For this interaction diagram The set of merged superpixel partitions in is the set of superpixel partitions The Superpixel partitions, , is the set of superpixel partitions The number of superpixel partitions in Partitioning into superpixels The sum of the weights of all edges connecting all pixels in with other superpixel partitions, For interactive diagrams The volume of Partitioning into superpixels The volume of Partitioning into superpixels The The sum of the weights of all edges connecting a pixel to other superpixel partitions.

8. The point cloud data segmentation method based on structural information theory according to claim 6, characterized in that: Calculating the reduction of the high-dimensional structure entropy relative to the current high-dimensional structure entropy includes: according to , calculating the reduction of the high-dimensional structure entropy relative to the current high-dimensional structure entropy; in, Partition the current superpixel The target superpixel partition directly adjacent to the current superpixel partition The reduction of the high-dimensional structure entropy after the merger relative to the current high-dimensional structure entropy, is the current high-dimensional structure entropy, Partition the current superpixel The target superpixel partition directly adjacent to the current superpixel partition The high-dimensional structural entropy after merging, For interactive diagrams The volume of Partitioning into superpixels The volume of Partitioning into superpixels The sum of the weights of all edges connecting all pixels in with other superpixel partitions, Partition the target superpixel The volume of Partition the target superpixel The sum of the weights of all edges connecting all pixels in with other superpixel partitions, Partitioning into superpixels Partition with target superpixel The combined volume, Partitioning into superpixels Partition with target superpixel The sum of the weights of all edges connecting all merged pixels to other superpixel partitions.

9. A terminal, characterized in that: The method comprises a memory and a processor, wherein the memory is used to store a computer program, and the processor is used to call and run the computer program stored in the memory to execute the method according to any one of claims 1 to 8.

10. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the steps of the method as claimed in any one of claims 1 to 8 are implemented.

Images