Abstract: In the welding process of gas metal welding (GMAW), due to the serious arc interference, it is difficult for the vision system to accurately extract the weld and the wire tip at the same time, thus affecting the accuracy of the weld tracking. An approach was proposed to locate the welding torch center instead of the welding wire tip. The feasibility of the method was demonstrated. First, after enhancing the weld seam and weld gun edge contour information in the molten pool image, a rectangular window was set to obtain the edge sampling point. Then, the clustering algorithm was used to screen out the correct edge sampling points. The weld line and the ellipse equation of the torch were fitted by the least squares method according to the sampling points. Moreover, the distance between the center of the current image welding torch and the straight line of the weld was calculated. Compared with the corresponding distance in the reference image, the amount of deviation of the welding torch position and the deviation of the welding gun swing were detected. The actual verification results show that the replacement error between the center of the welding torch and the tip of the welding wire is within 0.2 mm, which meets the requirements of tracking accuracy and has strong engineering practical significance.