Abstract: Objective To address issues such as reduced weld recognition accuracy and target drift caused by arc light and spatter during active vision automatic welding with line lasers, this study proposes an improved angle weld path recognition method based on correlation filtering. Methods Firstly, images of weld with laser stripes were captured by an industrial camera. Noise was removed via median filtering in scale space. The centerline of laser stripes was then obtained with the gray-level centroid method, and positions of weld feature points were determined through Hough line fitting. Then, correlation filtering tracks weld feature region, with reliability space weighting enhancing tracking accuracy. A local linear Kalman filter performed prior estimation of weld position, mitigating target drift during tracking. Finally, a dual-threshold strategy based on the average peak correlation energy enabled adaptive template updating, while image weighting fusion removed redundant information during template updates. Results Experiments demonstrate that the method controls recognition errors within 0.094 mm. Conclusion Compared to traditional correlation filtering, the proposed algorithm exhibits higher robustness and greater precision in identifying weld feature points.