Abstract: Compared with the on-line detection in production, traditional detection systems of camshafts' surface defect are mostly sampling after production in which missing rates and deficiencies rates are increased that can not be discovered until detection is completed, while full manually online detection not only increases the cost but also challenges the human resources. Therefore it is an important research subject to realize automatic real-time online detection. In this paper, a detection system of surface defect based on computer vision was designed. In this system, certain illuminants were constructed on both sides of the production line which surface images were captured by a high-speed camera and decided and located by an industrial computer while camshafts moved, stopped and rotated. For processing typical defect such as pit, pinhole porosity and rough surface, a defect segmentation algorithm and defect area marker algorithm were designed on the basis of defect features of camshaft surface. Target defect area was extracted accurately by the algorithms, defect locations were marked as well as the defect features were computed for final judgments. In this system, defect in camshaft surface which diameter is greater than 1 mm can be detected at 0.44 s/axis and locations of detect were displayed through human-machine interactive interface. Traditional sampling detection after production and full manually online detection can be completely replaced, meanwhile, the accuracy and efficiency are improved.