Visual method for weld seam recognition based on multi-feature extraction and information fusion

The complexity of welding process and the diversity of recognition object require high stability and reliability of weld seam recognition and tracking technology. These requirements can hardly be satisfied by recognition through single feature. An information fusion method based on confidence coefficient weighted probability addition is proposed: the shape feature of structured light stripe and the gray-level feature are extracted from weld seam images captured under different illumination situations. The edges of weld seam are recognized based on the two feature values,represented by probability densities of the edge and the confidence coefficients. The information fusion is realized by weighted probability addition of the probability densities. Seam recognition cases are studied to demonstrate the feasibility of the fusion algorithm. The result demonstrates that the algorithm performs well in terms of recognizing the weld seam edges with precision and stability.