A real-time welding seam detection technology based on laser vision is proposed to improve the speed and accuracy of welding seam detection.In the actual welding process,the collection of welding images has been a complicated process due to a large amount of noise interference.In this paper,a detection system based on laser vision is established to obtain the original image of laser stripe.On this basis,the original laser stripe image is grayed,and then an improved fast median filter algorithm is proposed to remove the salt and pepper noise in the image and shorten the program running time.Then the binary image of the laser stripe was obtained by Otsu threshold segmentation to extract the region of interest in the laser stripe.Then,the center line of laser stripe was extracted by combining the direction template method and ridge line tracking method,and finally the feature points of welding seam were extracted by sub-pixel corner detection method.Experimental results show that this method overcomes the influence of welding environment effectively,not only shortens the detection time of feature points of welding seams,but also has a higher detection accuracy and meets the actual welding requirements.