ClassiÞcation of human body motion

The classiÞcation of human body motion is a difÞcult problem. In particular, the automatic segmentation of sequences containing more than one class of motion is challenging. An effective approach is to use mixed discrete/continuous states to couple perception with classiÞcation. A spline contour is used to track the outline of the person. We show that for a quasi-periodic human body motion, an autoregressive process is a suitable model for the contour dynamics. This can then be used as a dynamical model for mixed state CONDENSATION Þltering, switching automatically between different motion classes. We have developed “Partial Importance Sampling” to enhance the efÞciency of the mixed state CONDENSATION Þlter. It is also shown here that the importance sampling can be done in linear time, in place of the previous quadratic algorithm. ”Tying” of discrete states is used to obtain further efÞciency improvements. Automatic segmentation is demonstrated on video sequences of aerobic exercises. Performance is promising, but there remains a residual misclassiÞcation rate and possible explanations for this are discussed.