A 3D Sonar Point Cloud Image Segmentation Method Based on Single Gaussian Model

Abstract

The embodiment of the invention provides a single Gaussians model-based three-dimensional sonar point cloud image segmentation method. The method is characterized by comprising the steps of (1) acquiring sonar data, and converting a three-dimensional sonar depth image corresponding to each frame of the sonar data into point cloud data under an orthogonal coordinate system; (2) sampling the point cloud data, taking a variable resolution R as a parameter, and down-sampling the point cloud data into a plurality of adjacent voxels with the length thereof to be R; (3) carrying out image segmentation on down-sampled voxels containing points by utilizing a single Gaussian model; (4) carrying out accumulation evaluation on each of all voxels by adopting the above image segmentation result, and separating foreground data from background data according to a cumulative score; (5) carrying out threshold filtering on down-sampled voxels based on the average value of n frames of pixel intensities; (6) for each voxel, if the result of the filtering treatment contains background data obtained in the step (4), removing the background data, and taking the remaining result as final foreground data.

Description

technical field [0001] The invention belongs to the field of three-dimensional sonar point cloud image processing, and in particular relates to a three-dimensional sonar point cloud image segmentation method based on a single Gaussian model. Background technique [0002] The phased array three-dimensional sonar system is a new type of underwater three-dimensional acoustic imaging system, which has wide application value in real-time monitoring of ports and underwater operation protection of frogmen. [0003] However, when obtaining 3D sonar point cloud data, due to the influence of equipment accuracy, operator experience, underwater environmental factors, etc., as well as the influence of the surface properties of the measured object and the operation process of data splicing and registration, the point cloud data Some noise spots will inevitably appear. In practical applications, in addition to the noise points generated by these measurement random errors, due to the influ...

Images