An appearance-based body model for multiple people tracking

摘要

We describe an appearance-based human body model for trackingmultiple people when they are interaction with each others causingsignificant occlusion amongst them, or when they re-enter the scene. Theproposed model allows real time surveillance systems understand“who is who” after multiple people interactions, partial ortotal occlusions, or when a person “reappeal” in the scene.It combines the grayscale textural appearance and expected shapeinformation together in a 2D dynamic template. Experimental resultsdemonstrates the robustness and real time performance of the proposedmodel