Goal

There is a huge demand for the automation of object tracking and event detection in video streams. While the number of cameras in our environment is constantly increasing, their use usually remains limited because currently there are no reliable software solutions that allow robust, real-time tracking of crowds in public places such as airports, train stations or banks. Similarly, the instantaneous, automatic annotation of sports event broadcasts such as soccer games is currently a dream. Current solutions (where they exist) require the deployment of electronic devices on each target to be tracked. This may be possible in the sports case, but is not in surveillance of public places.

A typical example is a world cup soccer match where up to 18 cameras are used to record the scene. Players form lumps and occlude each other. Our challenge is to integrate the information from all available cameras into a coherent representation, without noticeable delay, at high metric accuracy, while the cameras pan, tilt and zoom, covering ever-changing parts of the scene.

The TRICTRAC project aims at providing real-time object tracking and scene interpretation algorithms in difficult, dynamic scenes such as a soccer match or a crowd at an airport. It will provide base technology for a variety of video-analysis applications such as automatic annotation of sports broadcasts (detection of off-sides), surveillance (detection of suspicious behavior), and others.