Demo paper: A depth-based crowded heads detection system through a freely-located camera

This work demonstrates a novel system for locating multiple human subjects in a tight crowd by detecting their heads. By exploiting the 3D point cloud captured from a fixed depth camera, our system first develops a user interface for the users to interactively adjust the observation view to a proper place for head detection. Next, the proposed probabilistic framework can detect the human heads in the depth images from the manually adjusted bird´s eye view camera. The proposed interactive head detection system can provide high detection rate and real-time execution. Even a crowd of people are staying together shoulder-by-shoulder, the proposed system is still able to give promising people detection capability.