Local Feature Based Person Detection and Tracking Beyond the Visible Spectrum

摘要: One challenging field in computer vision is the automatic detection and tracking of objects image sequences. Promising performance local features feature based object approaches visible spectrum encourage application same principles to data beyond spectrum. Since these dedicated detectors neither make assumptions on a static background nor stationary camera, it reasonable use as basis for tasks well. In this work, we address two introduce an integrated approach both challenges that combines bottom-up tracking-by-detection techniques with top-down model strategies level features. By combination single framework, achieve (i) identity preservation tracking, (ii) stabilization detection, (iii) reduction false alarms by verification results every step (iv) through short term occlusions without additional treatment situations. our solely works independently underlying video-data specifics like color information—making applicable both, infrared data. detector trainable methodology does not any class specifics, overall general class. We apply task person For case show inherently allows component classification, i.e., body part detection. To usability approach, evaluate different real world scenarios, including urban scenarios where camera mounted moving vehicle.