Real-time pedestrian mapping from camera view to world space
Madhawa Vidanapathirana, Imesha Sudasingha, Pasindu Kanchana, Jayan Vidanapathirana, Indika N. Perera · 2017
Our environment is monitored by increasing number of surveillance cameras every day. Today, real-time automated analysis of surveillance video footages is a growing trend. These algorithms detect, track and analyze humans in videos and generate useful information. However, an analysis undertaken in camera view space yields only a limited amount of information that is relative to the camera view or screen space. We propose a technique that could be used to transform the screen space person detection to world space. Coordinate points generated on world space are independent of camera positions and thus could be used to combine information from multiple cameras. Additionally, such information can be combined with floor planning of considered area to generate additional useful information. Through rigorous testing and tuning, we have reached accuracy within 17 cm or approximately 1/3 the radius of personal space consumed by a person in determining the world space mapping.