Different shapes of parts have individual characteristics. In order to further study the accurate positioning of specific shape parts, this paper firstly extracts the straight line and circle features in the contour of a given part by Hough transform, and then uses the peak detection and K-means cluster analysis method to be accurate. Identify effective line and circle features; establish the absolute coordinate system of the part contour and the part standard position template coordinate system in turn; extract the effective pixel points of the part contour based on the full feature of the part and the main features of the part, respectively, and pass the simulated annealing algorithm The speed and accuracy of identifying the position of the part in these two ways are solved for different iterations. In order to improve the recognition speed, a fast recognition model based on the geometric features of the part is established, which is to simplify the objective function to the shortest distance and minimum of the center of the tested part and the center of the standard position template, and solve it by simulated annealing algorithm. The results show that the recognition time of part is 0.321s, the lowest recognition accuracy is 98.7%. It is included that the method has high recognition accuracy and fast recognition speed.