Abstract: During the process of GMAW (gas metal arc welding) seam tracking for the heavy thickness steel with the groove of V-shape, the visual features were degraded and the welding seam tracking accuracy was decreased under the effect of strong arc light and metal spatter interference. As a result, a lightweight dual task YOLO11n-seg (DT-YOLO) weld seam tracking method based on passive vision was proposed. This method used attention enhancement backbone to suppress noise, and it parallelly located the assembly gap and the welding wire in a single inference. Then RANSAC and weighted least squares methods were used to fit the two types of center points into a straight line. The offset of the welding gun was calculated by using the signed distance of two fitting lines along the normal direction. Finally, the lightweight deployment and temporal smoothing were combined to stabilize control signals. The results showed that under the V-shaped groove condition, the end-to-end delay of the system was about 50 ms, the closed-loop correction MAE was 0.067 mm, and the maximum error was ≤ 0.10 mm. This method is high robustness and low cost without an active light source, which is suitable for online tracking and real-time correction of industrial GMAW process. It has a good engineering application prospect.