The traditional iterative closest point registration method relies on the initial pose of the registration point cloud, otherwise it will fall into the local optimal solution during iteration, resulting in low registration accuracy. Aiming at the shortcomings of the traditional iterative closest point registration method, this paper proposes a point cloud registration method based on the eigenvalue deviation ratio, so that the point cloud can have a better initial pose when iterative closest point registration. The point cloud registration method proposed in this paper firstly performs voxel filtering preprocessing on the decentralized initial point cloud. Secondly, the method uses the deviation ratio of the eigenvalue to screen out the feature points. Thirdly it uses the invariant characteristics of the neighbor dimension of the point cloud to find the matching points of the two groups of point clouds, and then performs coarse registration. Finally the method uses the improved iterative closest point for fine registration. Experimental results show that the point cloud registration method proposed in this paper can make the registration point cloud have a good initial pose, compared with iterative closest point registration methods and registration methods based on sampling consistency initial registration and normal distribution transformation fusion. The proposed method in this paper has greater efficiency and precision.