Nowadays a large amount of manpower and material resources are required for mechanical reading，and thus an automatic mechanical watch accuracy measurement method based on machine vision is proposed in this paper. Firstly，a hardware system for image acquisition is designed，and then the improved mask-rcnn network is used to extract the mechanical pointer area and other area on the captured image. Secondly，the canny operator is used to extract the edge of the dial，and the centroid of each connected domain is determined. According to the distance relationship between the centroids and the midpoint of the line between the 12 o′clock and the 6 o'clock，the scale centroid is selected and used to fit the inner circle of the dial to determine the position of the center of the dial. Subsequently，the connection between the center and the center of 12 o′clock is used as the reference line for measurement，and the center line of each pointer is then extracted by the RANSAC least squares method. By calculating the angular relationship between the reference and center line，the purpose of using the machine to read the time is achieved. At the same time，as for the reading of the mechanical watch，the error correction principle is proposed，which greatly improves the efficiency of the mechanical watch accuracy measurement.