A Local-motion-based probabilistic model for visual tracking 1 M. Kristan b,a,1 , J. Perˇ s b , S. Kovaˇ ciˇ c b , A. Leonardis a a Faculty of Computer and Information Science, University of Ljubljana, Slovenia b Faculty of Electrical Engineering, University of Ljubljana, Slovenia Abstract Color-based tracking is prone to failure in situations where visually similar tar- gets are moving in a close proximity or occlude each other. To deal with the am- biguities in the visual information, we propose an additional color-independent visual model based on the target’s local motion. This model is calculated from the optical flow induced by the target in consecutive images. By modifying a color-based particle filter to account for the target’s local motion, the com- bined color/local-motion-based tracker is constructed. We compare the com- bined tracker to a purely color-based tracker on a challenging dataset from hand tracking, surveillance and sports. The experiments show that the proposed local-motion model largely resolves situations when the target is occluded by, or moves in front of, a visually similar object. Key words: Local-motion, Probabilistic visual models, Visual tracking, Occlusion PACS: : 1. Introduction In recent years, particle filters [1] have become a popular approach to track- ing from video due to their ability to efficiently handle the uncertainties associ- ated with the visual data and the target’s dynamics. Probabilistic trackers such as particle filters usually use contour-based [2] or color-based [3, 4] appearance models to locate and track the target. One drawback of these models is that tracking may fail whenever the target gets in close proximity of another visually similar object. In many applications, such as video surveillance, visual human- computer interface and tracking in sports, the camera is often positioned such 1 Accepted to Pattern Recognition Journal, 5 January 2009, Available online 10 January 2009, doi:10.1016/j.patcog.2009.01.002 * Corresponding author. URL: http://www.vicos.uni-lj.si (M. Kristan) Preprint submitted to Elsevier July 17, 2009