Abstract: Due to the constraints of natural lighting and other conditions, the development of weld seam extraction algorithms based on passive vision sensors is relatively slow. Aiming at this problem, this paper proposes a welding seam extraction algorithm based on intrinsic image decomposition. Firstly, the noise in the original weld image is suppressed by preprocessing, and secondly, the preprocessed image is decomposed into foreground and background layers based on gradient sparse prior. Among them, the foreground layer only contains structural information such as image edges, while the background layer contains smooth information such as lighting. Finally, Gamma transform is performed on the foreground layer to enhance the characteristics of the weld area and weaken the influence of lighting conditions on the algorithm. The experimental results show that the algorithm meets the requirements of the accuracy and real-time performance, and achieves the purpose of assisting the wall-climbing robot to locate the weld position.