Abstract: Precise initial positioning guidance in robotic welding is a key factor in ensuring welding quality. To address the issue of initial positioning deviations caused by premature detection in single-line laser vision sensors, this paper proposes a high-precision guidance method based on parallel dual-line laser vision. In this study, a self-designed vision sensing system is developed, with a core component being a parallel dual-line laser sensor integrated with a narrow-band filter (center wavelength 660 nm) and an attenuator to effectively suppress welding arc interference. The innovation of the method lies in a fast calculation algorithm based on a single-frame image: when the sensor moves to the point where the first laser line breaks at the starting edge of the workpiece while the second laser line remains intact, the system extracts seven key feature points from this image and, combined with precise hand-eye calibration and coordinate transformation relations, directly calculates the three-dimensional coordinates of the welding start point in the robot base coordinate system. Experimental results on the FANUC robot platform indicate that this method does not require performing a complex full-length weld scan. It can achieve rapid and precise initial position positioning through a single image capture, effectively compensating for the premature detection errors of traditional methods, and enhancing the accuracy and efficiency of the robot's initial welding positioning.