Collect point cloud data of objects with commonly used 3D scanning equipment, the resulting point cloud data is huge. Traditional point cloud registration algorithms cannot guarantee both efficiency and accuracy. To this end, combining octree-based K-means clustering point cloud streamlining algorithm with an ICP algorithm with improved weight ratio. Clustering using K-values of octree solid leaf nodes as mean clustering algorithm and initial clustering center, Calculate the root mean square curvature of each point and the average of the root mean square curvature of all points, the average Euclidean distance from each point to the cluster center and the average Euclidean distance from each point to the cluster center, and use this as a streamlined point cloud Data standards. Add a set of iteration method with variable weights during each iteration of the ICP algorithm to enhance the robustness of the algorithm, Experimental results show that the resistance of the algorithm is very robust, and it improves the defects of compound iteration and improves the registration efficiency.