Crowd Density Estimation from Autonomous Drones Using Deep Learning: Challenges and Applications

Abstract

Crowd flow estimation from Drones or normally referred as Unmanned Aerial Vehicle (UAV ) for crowd management and monitoring is an essential research problem for adaptive monitoring and controlling dynamic crowd gatherings. Various challenges exist in this context, i.e. variation in density, scale, brightness, height from UAV platform, occlusion and inefficient pose estimation. Currently, gathering of crowd is mostly monitored by Close Circuit Television (CCTV) cameras where various problems exist, i.e. coverage in little area and constant involvement of human to monitor crowd which encourage researchers to move towards deep learning and computer vision techniques to minimize the need of human operator and thus develop intelligent crowd counting techniques. Deep learning frameworks are promising for intelligent crowd analysis from frames of video despite the fact of various challenges for detecting humans from unstable UAV camera platforms. This research presents rigorous investigation and analysis in existing methods with their applications for crowd flow estimation from UAV. Besides, comprehensive performance evaluation for existing methods using recent deep learning frameworks is illustrated for crowd counting purposes. In addition, strong foundation for future direction is given by elaborating observations on existing research frameworks.