Image feature-based SLAM for flat surface modeling in indoor environment

In this study, we focused on image features in point cloud data to improve robustness of simultaneous localization and mapping (SLAM) for indoor mapping. We also focused on 3D area scanner to acquire point cloud data with reflection intensity images for image feature-based SLAM. Our methodology consists of five steps. First, point cloud data are acquired using a time-of-flight (TOF) camera from continuous viewpoints. Second, intensity images are generated from point cloud data. Third, the feature points are estimated from reflection intensity images with feature descriptors. Fourth, camera rotation matrices are estimated using corresponded feature points in intensity images. Finally, acquired point cloud data are registered using the estimated rotation matrices. We clarified that our methodology can integrate point cloud data successfully through our experiments in indoor environments.