Abstract: Concerned the problem of the closet search efficiency and registration accuracy difference in the process of three-dimensional point data registration, an improved algorithm of iterative closest point based on the unit quaternion is presented. At first, pairs of closet point are searched by using a strategy whose direction can be controlled for finding nearest neighbors of secondary rasterizing point cloud data. Then calculate the transformation matrix between reference point clouds and registration point clouds on the basis of pairs of closet point. Finally, on the basis of transformation matrix point cloud data registration is implemented. Simulated experiments showed that the provided algorithm could reduce registration time and increase registration accuracy to a certain degree which indicates that it could be more effective and feasible.