Abstract: Feature points extraction is the key to visual sensing and recognition of seam image. One of the most widelyused methods is to analyze slope of laser stripes,which is simple in principle and fast in computing speed. However,the method is indicated weak in adaptability and accuracy,even incapable for complex welding grooves. On the basis of exhaustive study on recognizing seam image from scanning-laser sensing system,a new idea called "feature from shaping" has been propounded that features points can be divided into inflection points at right angle and bevel angle based on the exact shape of seam groove. Both of two types of inflections are extracted by specially-designed algorithms,the former by method of extreme slope value and the latter by method of slope-intercept. During extraction in real-time seam tracking,corresponding algorithm is successively invoked according to the type of feature points until end of image processing. Experiments show that extracting feature points from shaping have advantages of high precision,strong anti-interference ability,good adaptability along with great practical value in the field of deep-groove seam tracking.