In the present paper, we propose an improved solid voxelization method based on Garcia and Ottersten’s method, which is known as CPU-based solid voxelization. When their method is applied to point cloud data sets obtained from consumer-level 3D cameras, such as Kinect and RealSense, parts of the surface completion may fail. This was found to be caused by certain adjacent voxel pairs generated in their surface voxelization step. In order to avoid such flawed voxel pairs, we propose an improved algorithm that uses an automatic rejection mechanism of flawed pairs and multiple shell completion. We apply the proposed algorithm to practical point cloud data sets and demonstrate the effectiveness of the algorithm in comparison with the original algorithm proposed by Garcia and Ottersten.

solid voxelization, surface voxelization, point cloud, RGB-D camera, depth image.