Accurate moving object detection in thermal imagery

Accurate detection of moving object provides a fundamental capability that drives numerous high-level computer vision applications. In this paper, a novel algorithm is proposed to detect objects in widely varying thermal imagery. On the basis of properties of thermal imagery, effective foreground and background model are first presented, and then these models are competitively used in an unified MAP-MRF framework to detect objects. Experimental results on challenging dataset show the robustness and effectivity of the proposed algorithm.