Abstract: Most of the weld tracking image processing algorithms have redundant operations in the filtering process and have poor real-time problems. In order to reduce the large number of redundant operations in the process of filtering noise in the weld image, based on the morphological characteristics of arc light splashes and laser stripes in the weld image, the first-order Markov chain was used to describe the spatial correlation of image matrix neighborhood, and a laser welding seam tracking image processing algorithm based on morphological feature filtering was designed. When the centerline of the laser fringe is extracted, this algorithm eliminates the noise interference in the welding image through the determination of the morphological features, effectively reduces the redundant operation of the filtering process, and improves the real-time performance. The experimental results show that the method has strong anti-interference ability, can accurately extract the center line of laser fringe, and it takes about 20 ms to process the entire image with a resolution of 576×768. It can complete real-time tracking welding with an absolute tracking error of 0.1358 mm.