Abstract: Seam tracking is one of key technologies in robotic automatic welding. To implement autonomous seam tracking, the key is to implement accurate extraction of laser stripes used to detect weld seam profiles in thick plate welding. According to the case where weld pools and laser stripes are simultaneously in the same frame, a method to effectively extract laser stripes was proposed based on the comprehensive orientation feature maps resulting from multi-region and multi-orientation Gabor filtering. Local-region average gray maximum method was used to remove interference, which is followed by connected domain methods to remove much small interference data. Experimental results show the effectiveness and it lays a good foundation for real-time seam tracking.