Real-time acquisition and representation of 3D environmental data

We develop a mobile system to acquire and represent 3D environmental data for modeling indoor spaces. The system is composed of a laser range finder (LRF) and an omni-directional camera. Multiple 3D point clouds from different viewpoints are acquired as the geometric information by scanning a scene with the LRF, while an omni-directional texture image is acquired with the omni-directional camera. We merge those multiple 3D point clouds into a single point cloud. We then combine the point cloud and the texture image into a complete 3D mesh model in three steps. First, we down-sample the point cloud based on a voxel grid and estimate the normal vector of each point. Second, using the normal vectors, we reconstruct a 3D mesh based on the Poisson surface reconstruction. Third, to assign texture information to the mesh surface, we estimate the matching region in the omni-directional image that corresponds to each face of the mesh. Simulation results demonstrate that the proposed system can reconstruct indoor spaces effectively.