Stereo vision system for welding line tracking robot

This paper describes a stereo vision system for welding line tracking robot which is designed for diagnosis of industrial pipelines and reservoirs. With the non-destructive test for checking internal crack of iron structure, vision system usually inspects the outside structural corrosion conditions. The stereo vision system in our research is mainly used for finding 3D directional information to guide mobile robot and for reconstructing the 3D shape of welding lines. The two image data taken from two different positioned cameras are transmitted via wireless communication and then used to compute 3D information. However, stereo vision process is yet far from solved clearly since there are many unexpected conditions such as noise, lighting difference between two images, quantization error, acquisition error, computational demand, and so forth. In this paper, the phases of two input images are used to reduce the above mentioned errors with the assumption of epipolar lines and parallel image planes. Computing 3D shape from two 2D phase images is then converted to a problem of finding the minima of energy function defined as MAP estimate. Experimental result shows good reconstructed 3D visualized welding lines with proper directional information.