A disparity map segmentation algorithm for 3D reconstruction of weld workpiece

In remote welding, the segmentation of disparity map is an important step to create 3D model of weld workpiece by stereo vision sensor. In this paper, the USF plane range image segmentation algorithm was introduced into disparity map segmentation, and by a region combination step, the revised algorithm can deal with the disparity map containing cylinder surface. The combination step could be divided into boundary detecting and curve region relabeling. During boundary detection, the pixel at the boundary of the segmented region after plane segmentation was recorded. In curve region relabeling, the adjacent regions in the same curved surface were assigned with the same label by comparing the boundary pixels'normal direction and distance between them. A segmented result of disparity map of saddle workpiece is shown to prove the feasibility of the algorithm.