Inferring Human Upper Body Motion Using Belief Propagation

Abstract

We present an algorithm for automatic inference of human upper body motion in a natural scene. A graph model is proposed for inferring human upper body motion, and motion inference is posed as a mapping problem between state nodes in the graph model and features in image patches. A multiple-frame inference algorithm is proposed to combine both structural and temporal constraints in human upper body motion. Belief propagation and dynamic programming algorithms are utilized for Bayesian inference in this graph. We also present a method for capturing constraints of human body configuration under different view angles. The algorithm is applied in a prototype system that can automatically detect upper body motion from videos, without manual initialization of body parts. We present evaluation results of our algorithm.