Bare Bones Strategy for Human Detection and Tracking

We present a scaled down version of a human detection and tracking system designed to run on relatively low-end machines of developing countries. The system uses sequence of monocular images of a single, fixed surveillance camera to extract data of moving objects. A linear motion model coupled with pre-computed average color intensities is used to track the detected subjects across a series of frames. Head detection algorithm is applied to form multiple hypotheses so as to accurately detect individuals in case of occlusion caused by people overlapping with each other. A final shape fitting algorithm is applied on the detected form to verify each hypothesis. Experiments conducted on real world data show the robustness of the algorithm, the speed of the process and its potential in lightweight, economical realtime applications