Fast Tracking of Humans in Frequently Occurring Entry, Exit and Occlusion Scenarios

Tracking problem can be formulated as the task of recovering the spatio-temporal trajectories for an unknown number of objects appearing and disappearing at arbitrary times. This work describes a modified mean shift clustering method for object detection. A human tracker based on the inter frame displacements of detected objects is proposed, where two different human classifiers based on size of detected clusters are used to handle different tracking issues. Human is separated from an occlusion group based on the information of direction of movement. Detection and tracking results are demonstrated and compared with results obtained using mean shift mode seeking approach. Results show that the proposed tracker is fast and reliable in situations where frequent entry, exit and occlusion of human are happening.