Visual tracking for video surveillance and vehicle safety

Övrigt konferensbidrag, 2010

This paper proposes a robust and efficient visual tracking method from videos. Potential applications, among others, include traffic safety for vehicles and freights, for video surveillance in airports, schools and banks. The proposed tracking scheme exploits both local point features and global appearance distributions of target objects. A novel online learning method is also employed to dynamically update the local features and global distributions of nonstationary visual objects. Our experimental results have demonstrated that the proposed scheme is very robust in tracking visual objects through complex video scenarios containing long-term partial occlusions, object intersections, severely deformed objects, objects with pose changes, with fast and sudden motions, and with cluttered background. Comparisons with two existing state-of-the-art methods have shown marked improvement.