Importance Sampling Based Probabilistic EigenTracker

Tracking objects over the video frames finds many applications in surveillance [13], human activity analysis [4] and gesture recognition [6],[12]. It becomes a very challenging task if the appearence and the dynamics of the object varies over successive frames. It becomes important to include elements of nonlinearity and non-Gaussianity in order to model accurately the underlying dynamics of a physical system [1]. We use Condensation Algorithm over the constructed eigenspace as the observation for the particle filter. The other problem with the tracking based algorithms is loss of track after sometime and it requires reinitialization after every few frames. Hence the trackers fail to track the entity over a long duration of time. By using Importance framework [9] we are able to increase the successful tracking time by a considerable amount. We also use parameteric transformation which gives a much tighter bound on the object to be tracked.