The invention belongs to the technical field of computer vision, and particularly relates to a point cloud automatic registration method based on a local curved surface feature histogram. The method comprises the following steps: carrying out cyclic voxel filtering on a source point cloud, downsampling the source point cloud to a designated point number, and carrying out voxel filtering on a target point cloud according to the size of an obtained voxel; carrying out key point searching and feature description on the basis of points with the maximum neighborhood curvature mean value of pre-keypoints, detecting the points with the curvature larger than 0.02 and serve as the pre-key points, calculating the curvature mean value of the neighborhood points of the points, and classifying the points with the maximum local curved surface curvature mean value as the key points; calculating a feature descriptor of the local curved surface histogram according to the relationship between the center of gravity of the point cloud in the neighborhood of the key point and the normal and distance of each point in the neighborhood; and calculating a mutual corresponding relation between the source point cloud feature descriptor and the target point cloud feature descriptor, removing an error corresponding relation according to random sampling consistency, and estimating an optimal coordinate transformation matrix according to SVD. Application prospect is wide.