# US 11,443,437 B2 · Vibe-based three-dimensional sonar point cloud image segmentation method > Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt - Number: US 11,443,437 B2 - Document ID: US11443437B2 - Status: Granted - Kind: B2 - Application number: 16/980,390 - Filed: 2018-10-22 - Published: 2021-02-18 - Granted: 2022-09-13 - Earliest claimed: 2018-03-14 - Assignees: Zhejiang University - Applicants: Zhejiang University - Inventors: Yaowu Chen; Xiang Tian; Xuesong Liu; Rongxin Jiang - Web page: https://patentgrep.com/patent/US11443437B2 ## CPC - G06T 7/136: involving thresholding - G06T 7/11: Region-based segmentation - G06T 7/187: involving region growing; involving region merging; involving connected component labelling - G06T 7/194: involving foreground-background segmentation - G06T 2207/10028: Range image; Depth image; 3D point clouds - G06V 10/28: Quantising the image, e.g. histogram thresholding for discrimination between background and foreground patterns - G06V 20/64: Three-dimensional [3D] objects - G06T: IMAGE DATA PROCESSING OR GENERATION, IN GENERAL - G06V: IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING ## Abstract The present invention discloses a ViBe-based three-dimensional sonar point cloud image segmentation method, characterized by including: (1) acquiring sonar data, and converting three-dimensional sonar depth image data corresponding to each frame of the sonar data to point cloud data under an orthogonal coordinate system; (2) sampling the point cloud data, and down-sampling the point cloud data to a plurality of adjacent voxels with side length being R by taking a variable resolution R as a function; (3) carrying out image segmentation on the down-sampled point cloud data by a ViBe algorithm; (4) carrying out accumulative scoring on each voxel according to an image segmentation result, and sorting foreground data and background data according to accumulative score; and (5) clustering the foreground data, and then carrying out expansion operation on the foreground data by taking an original point in the voxel as a center to obtain final foreground point cloud data. ## Claims ### Claim 1 A Visual Background Subtraction (ViBe) based three-dimensional sonar point cloud image segmentation method, characterized by comprising the following steps of: (1) acquiring sonar data, and converting three-dimensional sonar depth image data corresponding to each frame of the sonar data to point cloud data under an orthogonal coordinate system; (2) sampling the point cloud data, and down-sampling the point cloud data to a plurality of adjacent voxels with side length being R by taking a variable resolution R as a function; (3) carrying out image segmentation on the down-sampled point cloud data by a ViBe algorithm; (4) carrying out accumulative scoring on each voxel according to an image segmentation result; sorting foreground data and background data according to accumulative score; and (5) clustering the foreground data, and then carrying out expansion operation on the foreground data by taking an original point in the voxel as a center to obtain final foreground point cloud data. ### Claim 2 (depends on 1) The ViBe-based three-dimensional sonar point cloud image segmentation method according to claim 1, wherein total point cloud data is directly down-sampled to obtain a plurality of adjacent voxels with side length being R. ### Claim 3 (depends on 1) The ViBe-based three-dimensional sonar point cloud image segmentation method according to claim 1, wherein a part of point cloud data is sequentially and segmentally down-sampled to obtain a plurality of adjacent voxels with side length being R. ### Claim 4 (depends on 1) The ViBe-based three-dimensional sonar point cloud image segmentation method according to claim 1, wherein the step (3) comprises the following steps of: (3-1) for each voxel, randomly selecting n points as a sample space of the voxel in N-neighborhood of the voxel; (3-2) during segmenting the point cloud image of a certain frame, carrying out similarity counting on each voxel of the frame and a sample space of the same coordinate corresponding to a point cloud image of the frame, adding 1 to a fitting counter C1 when the intensity difference between the voxel and a point in the sample space is smaller than a threshold T1, considering the voxel as the background data if the whole sample space is compared and the C1 is larger than or equal to the threshold T2, otherwise considering the background data as pre-foreground data; (3-3) during segmenting each frame of point cloud image, changing the voxel to the background data if one foreground point is judged as the pre-foreground data at times exceeding a threshold T3; and (3-4) updating one point of the sample space of the corresponding coordinate if one voxel is the background data and occurs with probability of 1/Alpha, and updating one point of the sample space of one own neighborhood if the probability of 1/Alpha exits. ### Claim 5 (depends on 1) The ViBe-based three-dimensional sonar point cloud image segmentation method according to claim 1, wherein in the step (4), if one voxel, in the calculating process, is judged as the pre-foreground data at times not less than S1 times when the step (3) is ended, the voxel is judged as the foreground data, otherwise the voxel is judged as the background data. ### Claim 6 (depends on 4) The ViBe-based three-dimensional sonar point cloud image segmentation method according to claim 4, wherein a value of threshold T1 is 0.01-10,000.00, a value of the threshold T2 is from 1 to point numbers of the sample space, and a value of the threshold T3 is 10-100. ### Claim 7 (depends on 1) The ViBe-based three-dimensional sonar point cloud image segmentation method according to claim 1, wherein in the step (5), the nearest neighbor-clustering method is adopted to cluster the foreground data to obtain one or more voxel sets, and a distance threshold T4 of clustering is 1.0-20.0. ### Claim 8 (depends on 7) The ViBe-based three-dimensional sonar point cloud image segmentation method according to claim 7, wherein in the step (5), an expansion operation with expansion kernels being E1, E2, and E3 is carried out on the foreground data after clustered results are reflected onto the original point in the voxel, and values of E1, E2, and E3 are all odd numbers from 1 to 21. ### Claim 9 (depends on 1) The ViBe-based three-dimensional sonar point cloud image segmentation method according to claim 1, wherein in the step (5), an expansion operation with expansion kernels being E1, E2, and E3 is carried out on the foreground data after clustered results are reflected onto the original point in the voxel, and values of E1, E2, and E3 are all odd numbers from 1 to 21.