Abstract: The seam tracking system based on structured light vision is widely used in robot welding, and the key to ensure the quality of such welding is the accuracy of seam recognition. Traditional weld seam recognition algorithms can hardly solve the interference problem of strong welding noise, so a weld seam recognition algorithm based on improved active contour model was proposed. Firstly, in order to overcome the shortcomings of manually obtaining the initial contour by traditional active contour model, a basic segmentation algorithm to automatically extract the initial contour was designed.Then, according to the grayscale distribution characteristics of the weld image with strong noise, the algorithm was used to optimize the energy function of the traditional active contour model, and the extended structure tensor was added to the external energy term to ensure the stability and accuracy of the convergence result. Finally, the weld feature points were extracted based on multi-segment line fitting and RANSAC line fitting. The results show that the algorithm has good adaptive accuracy under strong noise interference.