Cooperative estimation of human motion and surfaces using multiview videos
Computer Vision and Image Understanding
| Hi-index | 0.00 |
In this paper, we have developed a fast and accurate 3-dimensional registration algorithm using surface gradients for motion tracking of Time-Varying Meshes (TVM). Hierarchical skeleton-based mesh decomposition is used to segment each TVM frame into a kinetic model without user intervention. An iterative closest point matching algorithm with the use of vertex normal data is developed to resolve translational and rotational motion vectors. As only vertices with large surface gradients are used as matching features, processing speed can be reduced by 80% retaining a high level of accuracy. Multi-temporal registration, where historical and future data are used in the 3D shape registration, is used to increase robustness against potential bad registrations, achieving 50% error reduction.