Section: New Results
3D Bayesian Tracker
Participants : Augustin Lux [ correspondant ] , Rémi Emonet, James Crowley, Matthieu Langet, Jean-Pascal Mercier.
The 2DBT and 3DBT tracking systems are autonomic perceptual components created for the IST CAVIAR project and the IST CHIL project. Both systems are autonomous perceptual components managed by a autonomic supervisor. The Autonomic supervisor provides self monitoring, self repair, self configuration, auto-regulation of parameters and self-description.
The INRIA 3D Bayesian body tracker is used to detect, locate and track multiple 3D entities in a CHIL room in real time. It is configured and optimized for detecting and tracking people within CHIL rooms using multiple calibrated cameras. In theory the camera set can include an overhead panoramic camera, but as of this writing, the system has only been tested with wall mounted cameras, typically found in the upper corners of CHIL rooms. Cameras may be connected and disconnected while the component is running, but they must be pre-calibrated to a common room reference frame. The calibration data is obtained by reading a file obtained from the CHIL KBS.
This perceptual component can be configured to monitor and track the activity within a CHIL room. The tracker receives its observations from 2D detection process that can use any available pixel level detection algorithm. The tracker currently integrates information from adaptive background subtraction, motion detection, skin color detection, and local appearance using scale normalised Gaussian derivatives. A common scenario is to use the motion to detect and initialise tracking, adaptive background subtraction to track 3D bodies, and skin color to track hands and faces. Cameras may be connected dynamically.
The system 3DBT has been declared with the APP "Agence pour la Protection des Programmes" under the Interdeposit Digital number IDDN.FR.001.490023.000.S.P.2006.000.10000