Hybrid 3D Reconstruction: SLAM vs Homography-Based Motion Estimation

We present a hybrid framework for the acquisition of 3D textured models of objects of limited size. The framework combines active laser-based triangulation with passive motion estimation. The 3D models are obtained by imaged laser profile backprojection onto a time-varying frame and subsequent motion-based alignment with respect to a fixed frame. Two distinct techniques for estimating camera displacements are described and evaluated. The first is based on a Simultaneous Localization and Mapping (SLAM) approach, while the second exploits a planar pattern in the scene and recovers motion by homography decomposition. Results obtained with a custom laser-camera stereo setup — implemented with off-the-shelf hardware — show that the models can be reconstructed with a good quality, and that, in particular, a trade-off exists between operational flexibility and model accuracy.