Real-Time Tracking of Multiple People Using Continuous Detection

Recent investigations have shown the advantages of keepingmultiple hypotheses during visual tracking. In thispaper we explore an alternative method that keeps justa single hypothesis per tracked object for computationalefficiency, but displays robust performance and recoveryfrom error by employing continuous detection duringtracking. The method is implemented in the domain ofpeople-tracking, using a novel combination of stereo informationfor continuous detection and intensity image...