Feature-Based Monocular Dynamic 3D Object Reconstruction

Dynamic 3D object reconstruction becomes increasingly crucial to various intelligent applications. Most existing algorithms, in spite of the accurate performances, have the problems of high cost and complex computations. In this paper, we propose a novel framework for dynamic 3D object reconstruction with a single camera in an attempt to address this problem. The gist of the proposed approach is to reduce the reconstruction problem to a pose estimation problem. We reconstruct the whole object by estimating the poses of its topological segmentations. Experiments are undertaken to validate the effectiveness of the proposed method in comparison with several state-of-art methods.