Abstract: To address the problem of groove edge position detection accuracy degradation caused by interferences such as arc light, welding fume, and welding spatter in complex welding environments, a detection method for groove edge position based on probabilistic Hough transform was proposed. Passive visual sensing technology was adopted, and welding images synchronized with the arc position were acquired by a complementary metal-oxide-semiconductor (CMOS) camera equipped with a combined light-reducing and filtering system. By establishing a dynamic layered image processing method, the features of the welding groove edge were extracted, and the dynamic distribution patterns of welding interference data were analyzed. A probabilistic Hough transform-linear principal component (HT-LPC) analysis method was proposed to eliminate the interference data segments caused by interferences such as welding arc light, spatter, and fume. Based on the retained valid data segments, the groove edge position was detected. Narrow-gap double-layer single-pass welding tests indicate that the groove edge position detection accuracy of the HT-LPC algorithm is within ± 0.1 mm, providing technical support for the intelligent perception and control of the welding process.