Motion understanding from qualitative visual dynamics

In many situations, motion is naturally grasped as an ordered sequence of poses, expressing stages in a general development. Capturing the underlying dynamics is an important first step for gaining insight and constructing a representation that has expressive power. The authors compute a pose function for a motion that qualitatively characterizes it in terms of dynamics. The study is based in dynamic systems analysis, constructing the function in phase space-a geometric depiction of the observed behaviour. The analysis itself is geometrical, but yields a representation of the visual dynamics. The global component of the pose function is an expression of external forces, and the local component and expression of deformation. The computation is simple, making only few general assumptions. The function captures all the relevant parameters of the motion in the sense that a qualitative simulation of the behavior is made possible, thus, facilitating subsequent reasoning about observed events.<>