PROJECT TITLE :

Sequential Non-Rigid Structure from Motion Using Physical Priors

ABSTRACT:

We tend to propose a replacement approach to simultaneously recover camera cause and 3D form of non-rigid and probably extensible surfaces from a monocular image sequence. For this purpose, we tend to create use of the Extended Kalman Filter based Simultaneous Localization And Mapping (EKF-SLAM) formulation, a Bayesian optimization framework traditionally employed in mobile robotics for estimating camera create and reconstructing rigid situations. So as to increase the matter to a deformable domain we have a tendency to represent the item's surface mechanics by means that of Navier's equations, which are solved using a Finite Component Technique (FEM). With these main ingredients, we will more model the fabric's stretching, permitting us to go a step additional than most of current techniques, usually constrained to surfaces undergoing isometric deformations. We tend to extensively validate our approach in both real and synthetic experiments, and demonstrate its blessings with respect to competing ways. A lot of specifically, we show that besides simultaneously retrieving camera cause and non-rigid shape, our approach is adequate for both isometric and extensible surfaces, does not require neither batch processing all the frames nor tracking points over the full sequence and runs at many frames per second.