Automated indoor scene reconstruction with mobile robots based on 3D laser data and monocular visual odometry

This paper presents a method applied for automated 3D scene reconstruction for mobile robots. A monocular visual odometry based on ceiling vision collected from a digital camera is implemented to estimate the robot´s trajectory. To improve the capability of managing the large amount of data supplied by a 3D scanner, we adopt a fast scene voxelization algorithm where the 3D range data are stored in occupied voxel list. Moreover, local shape descriptors are calculated in a further step in voxel scale. For the purpose of fast and efficient convergence, a variant of ICP (Iterative Closest Point) algorithm is therefore proposed to accomplish multi-scenes registration by taking shape descriptors into account in correspondence phase. A series of experimental results demonstrate the performance of the proposed 3D indoor scene reconstruction method in an actual environment.