Abstract: A new method of monitoring weld formation by means of binocular stereo vision is put forward.First,the images of weld formation were collected from joints surface on the vision system.The images were pretreated for gray balance and Gauss filter before image segmentation.Second,edges of weld formation were precisely extracted from the binary images by means of dynamic threshold segmentation and morphological approach of expansion and corrosion.On the basis,a method to measure weld length,weld width,weld height and other weld formation parameters was explored.At last,the actual verification results showed that standard deviation of weld width s_W,standard deviation of weld height s_H,and transition angel α_L and α_R could be made as quantitative indicators of evaluating weld formation.