With the development of medical surgical navigation system, more and more orthopedic surgeries benefit from the accurate real-time positioning of the navigation system. Due to the extremely high time cost in surgery, how to achieve accurate and fast registration with a small number of manually extracted key points is a research problem in the field of medical point cloud registration. In this paper, the algorithm of point cloud rough registration stage is improved, and a new registration algorithm, SSAC+ICP, which is suitable for medical surgical navigation system with low overlap rate is proposed. Four groups of fusion experiments were performed to register the CT point cloud images of spinal contours and the manually collected key point cloud data. Experimental results show that the registration accuracy of SSAC algorithm is much higher than that of NDT, 4PCS and PCA.