Seam tracking technology is an important area of research automatic arc welding, precise seam tracking is crucial to achieve high quality welds. In order to achieve precise seam tracking, seam deviation (the weld center arc deviation) detection is a key. Unlike the conventional method by image processing techniques to obtain the seam deviation information directly, but selected image processing area (including the distal end portion of the molten pool welds and the front end of the pool), and analyzed as a pool image centroid characteristic parameters of the weld deviation. Study these parameters to create a new method for visual weld deviation measurement model, establish the weld center neural network model using artificial neural network modeling technology.