An Improved Supervoxel Clustering Algorithm of 3D Point Clouds for the Localization of Industrial Robots

Supervoxels have a widespread application of instance segmentation on account the merit providing highly approximate representation with fewer data. However, low accuracy, mainly caused by point cloud adhesion in localization industrial robots, is crucial issue. An improved bottom-up clustering method based supervoxels was proposed for better accuracy. Firstly, data were preprocessed to eliminate noise points and background. Then, supervoxel over-segmentation moving least squares (MLS) surface fitting employed segment clouds workpieces into clusters. Every cluster can be refined MLS fitting, which reduces occurrence that divides two objects patch. Additionally, an adaptive merging algorithm fusion features convexity judgment accomplish individual workpiece. experimental platform set up verify method. The results showed recognition accuracy rate three different kinds all over 0.980 0.935, respectively. Combined sample consensus initial alignment (SAC-IA) coarse registration iterative closest (ICP) fine registration, coarse-to-fine strategy adopted obtain location segmented experiments. demonstrate robots higher lower time.