Three-View Shape Recovery and Incremental Registration of 3D Point Sets

Abstract

This paper describes a shape recovery framework based on scaled orthographic factorization and incremental 3D point cloud registration. While the camera is moving freely in unknown environments, point features are automatically extracted and tracked. Using the tracked features in three consecutive frames, 3D locations are recovered and partial point subsets are immediately registered to the global point cloud. The advantage of the proposed incremental method is that the registration process does not rely on past frames for estimating the global 3D point cloud. The shape recovery and registration steps are combined as a single procedure that processes each input frame only once. All the internal steps are linear and non-iterative, hence the procedure is applicable to on-line interactive applications. Experimental results showing the feasibility of the presented approach are also provided.