Abstract: In the process of automatic welding, the installation error of the workpiece to be welded and the thermal deformation in the welding process lead to the weld deviation and affect the welding quality.Taking the plate lap welding seam as the tracking target, the welding seam images were collected through vision system and the weld position data was obtained to identify the welding seam and positioning.A method was proposed to correct the deviation of weld tracking data, and the method combining median filter and Kalman filter was used to correct the fluctuation of weld position data.The welding experiment shows that the welding seam can be identified and located accurately under the interference of solitary light and spatter.The proposed method can effectively filter the data fluctuation and improve the movement stability of the welding mechanism.