Sparse point clouds fitting in 3D reconstruction for welding environment

3D reconstruction for non-structured environment is the key technology in virtual environment in remote welding. The welding environment was reconstructed based on stereo vision. Because welding environment is lack of feature points for stereo matching. The tags were added onto welding workpieces to provide feature points. Sub-pixel corner extraction and human-assist stereo matching was applied to acquire sparse point clouds of welding environment. The least square algorithm for plane fitting and Tikhonov regulation for quadric surface were used to fit the sparse point. Quasi-optimality criterion was used to choose the regulation parameter. The workpieces of plane and cylinder were reconstructed and the 3D models of workpieces were obtained. The results show that the fitting deviations for both planar and quadric models are all less than 1 mm, and the maximum deviations are less than 3 mm.