High precision bearing is a kind of cylindrical parts. A machine vision based online detection system is designed and implemented to meet the need of high curvature surface defects detection of cylindrical parts. During the measurement, a special light source and a lighting system are used to solve the problem of the reflections on the metal surface. Through the combination of optical systems and mechanical rotating platform, cylindrical parts are imaged by the optical system in the process of rotation, then a complete cylindrical surface image can be acquired. Micron surface defects and dimensions of bearing can be detected after using the fast image processing technique. Through the analysis on the morphology of surface defects, the type of defects can be determined. Experiments show that the proposed system has the features of high efficiency, high accuracy and ease of use. It can effectively meet the needs of high precision bearing detection and dimension online detection.