Mutual information-based 3D object tracking

摘要

We propose a robust methodology for 3D model-based markerless tracking of textured objects in monocular image sequences. The technique is based on mutual information maximization, a widely known criterion for multimodal image registration, and employs an efficient multiresolution strategy in order to achieve robustness while keeping fast computational time, thus achieving near real-time performance for visual tracking of complex textured surfaces.