Small sample parts recognition and localization from unfocused images in precision assembly systems using relative entropy

Recognition and localization of mechanical parts using machine vision is a common approach in precision assembly systems. However, positional inaccuracy in assembly systems often produces unfocused images. Hence, existing methods of part recognition and localization are vulnerable to failure. In this paper, we present a part recognition and localization method, based on relative entropy, which can be applied to small samples. First, a template image is generated based on the contour of the parts and divided into several regions. The intensity distribution of the regions was sampled to generate template features. Then, the captured image is segmented using the Gaussian mixture model and the expectation maximum algorithm to extract the target part in the image. Part features are also generated by sampling the target part image using the template features. Furthermore, an optimization model is established in which the objective function is the sum of the relative entropy between the image features, the template features, and the region matching error correction term. By solving the optimization model, the location of the part can be obtained. The proposed method is compared with the edge and invariant feature-based methods through experiments. The results show that the proposed method has higher robustness and is suitable for the recognition and localization of parts with unfocused images. By using this method, the flexibility and reliability of precision assembly systems can be improved.