A new method for real-time rapid detection of arc-shape surface of polished metal based on machine vision is proposed. In the offline situation, the samples of artifacts under different light intensity are analyzed, the relation function of the background brightness component and the gray level is constructed, and the statistical characteristics of reflection component of the samples are extracted. Online testing, the image gray level and the corresponding brightness component are calculated. Then the reflection component whose gray level distribution is uniform is extracted from the image. Finally, the reflection component is processed with thresholding segmentation and decision is made. The experiments show that the proposed method can adapt to the changing light environment through one-time learning. The average time to detect an image is 40ms, and the accuracy rate is over 98%. Moreover, the system has high robustness, high real-time and high accuracy.