High Performance Optical Flow Serves Bayesian Filtering for Safe Human-Robot Cooperation

Abstract

Optical Flow estimation has found a way to be calculated in real time with feasible accuracy. This is due to novel extensions in the variational formulation and the progress developing modern graphic processor units (GPU) which enable parallel computations due to dozens of stream processors. This paper will show the development of a new data term for the variational formulation of the optical flow estimation and then embedding the optical flow field into a Bayesian filter for the purpose of human kinematic estimation. The theoretical considerations are manifested by experimental results. Novel theoretical considerations including an implemented framework with experimental results for markerless human motion and kinematic estimation.