Abstract: To improve the accuracy and operation efficiency of patient registration for the surgical navigation system， the registration strategy which combined three-point method and iterative closest point (ICP) algorithm was proposed. Firstly， the patient registration problem was defined and the acquisition methods of preoperative and intraoperative data were introduced. Then， the spherical center of the optical positioning markers were taken as the common features between patient space and image space， and the initial registration was completed by using the three-point method. Finally， the spherical regions with radius r were established by taking each point of patient point cloud after initial mapping as the spherical center， and the points of image point cloud located in these regions were retained to realize sampling， and the improved ICP algorithm was used to perform accurate registration of the two point clouds. The experimental results showed that the mean registration errors for pig femur and pig ilium when using the proposed method are (0.83±0.10)mm and (0.86±0.09)mm， respectively， which is superior to the traditional ICP algorithm in terms of accuracy and stability， and is efficient and easy to operate with certain potential values in clinical applications.

Key words: surgical navigation system, patient registration, sparse to dense point cloud registration, image point cloud sampling, ICP(iterative closest point) algorithm