Tracking articulated objects with physics engines

This paper presents a novel approach to track multiple articulated objects in a video sequence. The key idea is to define a model of the object using a set of geometrical primitives linked by physical constraints, and exploit physics engines to solve these constraints while the model adapts to the object under the influence of local mean-shift processes. This novel approach to object tracking has numerous advantages: the model provides rich geometric information about the object at the articulation level; multiple touching objects are implicitly distinguished using collision detection strategies; physics engines are able to efficicently manage both image-based and model-based constraints simultaneously for a neglectable computational cost, suggesting their potential interest for many more image processing applications.