Abstract: Addressing the challenge of achieving high-precision point cloud registration in 3D reconstruction using traditional coarse registration methods, this paper proposes a RANSAC algorithm enhanced through geometric constraint optimization and scale consistency. The proposed approach employed ISS and SHOT algorithms to identify feature point pairs, leveraging geometric constraint assessment between point pairs to eliminate outliers and enhance speed and accuracy of coarse registration. Furthermore, in response to the incapability of conventional methods to handle scale inconsistency in point clouds, the Umeyama algorithm was incorporated to calculate registration matrices by considering point cloud scale information, and the GICP algorithm was introduced to further enhance registration precision. Experimental results demonstrate that this algorithm has obvious advantages in registration tasks of point cloud models from various angles and effectively solves the problem of inconsistent point cloud scale.