A Review of 3D Pose Estimation from a Monocular Image Sequence

3D pose estimation is an important part of 3D motion estimation, which is useful for visualized operation and control. The paper classified approaches of 3D pose estimation from a monocular image sequence into three types: feature-based approach, optical-flow-based approach and model-based approach. In feature-based approaches, 3D motion are estimated by observing a set of 2D features, such as points, lines, junctions and corners, over two or more images. On the contrary, optical-flow-based approaches estimate instantaneous 3D motion from image plane velocity. Both of them can complete the estimation without any a priori knowledge of the object, but the computation is complex and time-consuming. Mode-based approach avoids the burden of computation, but relies on the a priori knowledge of the object. Current research status of each approach is reviewed. Comparison of these three approaches and developing trends of 3D pose estimation are discussed in the conclusion.