Multi-pose human head detection and tracking boosted by efficient human head validation using ellipse detection

In this paper, a novel system is developed to detect and track multiple heads in multi-pose status by efficient human head validation using ellipse detection. A particle filter is employed for each tracked head. The appearance model of head is updated based on fusion of color histogram and oriented gradient one. Regardless of its pose, a head is modeled as an ellipse, and we propose an objective function to fit the proposed elliptical equation. A robust supervised distance function learning framework has been developed to recover some missed detections and suppress some false detections, using an Expectation Maximization algorithm. Comparative study with state-of-the-arts has indicated the superiority and good performance of the proposed method.