Abstract: To solve the problems of low assembly flexibility and poor workpiece recognition robustness of industrial robot, a workpiece recognition method based on machine vision is proposed for parts assembly. First, the feature points are extracted uniformly using the double-layer rotating filter with dimensional invariance and rotational invariance, then the feature point descriptors with orientation are generated by the rBRIEF algorithm, and then the Hamming distance and RANSAC algorithms are used for feature point matching and optimization to accurately identify the target artifacts. The proposed method is tested with workpiece images, and the results of workpiece recognition by the method and other three traditional methods under rotation and scaling transformation are compared and found to be more accurate and efficient in recognizing workpieces. Finally, the experimental platform of shaft-hole assembly based on machine vision is built to verify the feasibility of workpiece recognition.