Abstract: To improve the intelligence level of narrow-gap welding, this paper proposes an adaptive localization method for the region of interest (ROI) window in groove edge images based on passive visual sensing. The method offers strong synchronization and robustness, effectively addressing the challenges of locating groove edges and achieving poor detection accuracy caused by variations in groove width during the welding process. By adjusting the horizontal translation of the groove edge ROI window, the influence of the ROI window position on the proportion of background to target in the groove edge image and the accuracy of edge feature extraction is revealed. A groove edge ROI window generation mechanism synchronized with the arc position is established, enabling the precise generation of groove edge ROI images and accurate extraction of groove edges during the welding process, thereby improving the applicability of narrow-gap welding technology. Variable groove welding tests further validate the adaptability and effectiveness of the proposed method in groove edge positioning and detection. The results show that when the horizontal translation of the ROI window is set to 50% of the ROI window width, the absolute deviation between the measured groove edge position and the mean position is less than 3.53 μm, and the average detection time is less than 30 ms per frame. These results meet the requirements for precision and real-time parameter detection in narrow-gap welding processes.