Chang Baohua

Research on a visual weld detection method based on invariant moment features

A visual weld trace detection method based on invariant moment features is proposed in this paper for cosmetic welding inspection. The proposed method can not only overcome the absence of distortion in structural light detection, but also solve the problems of existing detection method based on grayscale weld image such as many priori parameters needed, time-consuming feature extraction process, instable features etc. This research shows that the proposed detection method has advantages in better adaptability, fewer priori parameters and higher efficiency. The detection error of the proposed method is not more than 0.3 mm. The seam and base metal region can be classified effectively. The proposed method is expected to be applied to real-time seam tracking in multi-layer welding.