The gear structure is a key component of the transmission device, and the accurate detection of the addendum circle is an important basis for subsequent assembly. In the visual measurement of addendum circle, the traditional image processing method has low detection accuracy, and when the inclination angle of the gear structure is too large, the gear teeth will be occluded, which leads to the poor robustness of the algorithm. Aiming at the above problems, Detection method of addendum circle of gear structure based on machine vision is proposed. First, the sub-pixel corner detection of the gear teeth is performed based on the curvature scale space ( CSS) technology with adaptive threshold, second, the hyper least square method is used to fit the addendum ellipse, and finally the ellipse parameters are optimized by compensating for the quasi-eccentricity error. The experimental results show that the algorithm can not only extract the addendum circle that contains all the gear teeth images, but also can perform high-precision detection of the occluded images of the gear teeth. At the same time, it can compensate the elliptical quasi-eccentricity error caused by lens distortion. The measurement accuracy of the addendum circle center is 0. 056 mm, and the measurement accuracy of normal vector is 0. 068°, which meets the requirements of visual measurement of gear structure.