Laser-vision system (LVS), which provides the effective sensing of weld seams, is widely used in robotic welding. Due to the harsh visual noise during the welding process, the accurate real-time detection of weld seams is difficult. Since the optical propagation of the laser is affected by the arc light, the traditional passive vision system (PVS) filtering method for detecting weld seams is not applicable. This paper exploits the selection of an optimal imaging band for the LVS to eliminate the arc noise.