In order to solve the problems of measurement difficulty and precision in manual assembly mode of large-size gear structure with coaxial internal meshing, a visual measurement method for gear profile structure assembly is proposed. In this method, a monocular vision measurement system is constructed, and its spatial pose is calculated by using its image features to realize high-precision measurement. First, based on the minimum kernel value similarity area algorithm of adaptive kernel and adaptive threshold, the tooth tip angle points are extracted, and Canny algorithm improved by OSTU is used to detect discrete arcs of auxiliary circle. Then, combined with the idea of random sampling consensus, the ellipse fitting algorithm based on geometric distance and Levenberg-Marquardt optimization is used to obtain the ellipse parameters, and the measurement model of gear profile structure and the principle of pose calculation are given. Finally, the experiment is verified, and the experimental results show that the measurement accuracy of this method can reach 0.050 mm, which meets the requirements of high-precision measurement of gear structure assembly.