Tracking of pedestrians using a system of agile stereo cameras
Louis Buteau-Vaillancourt
Denis Laurendeau (Supervisor)
Problem: Our society (both locally and internationally) faces a larger diversity of attacks on our properties (buildings and people). We are searching for new ways to protect property and increase security. One avenue of research which is attracting attention is computer vision automated surveillance. Currently available systems suffer from the inability of cameras presently in place to extract precise 3D data as well as track subjects of interest, and thus a more detailed analysis and study of this problem is required.
Motivation: This study is a continuation of the MONNET project which is nearing completion, and will add a node formed by a pair of agile stereo cameras. The MONNET project is already able to extract information from an observed scene, but the system is very static. With the addition of a stereo pair, the system will be less static and will be able to react to its environment.
Approach: The approach involves detecting a subject in the field of view of the cameras and to extract the maximum amount of information while viewing the common field of view. When the subject leaves the common field of view, the system estimates the trajectory of the subject and the current position so as to redirect itself towards the subject and continue the tracking. This will be continued until the limit of the cameras’ movement is reached.
Challenges: One of the major challenges of this project is background subtraction. Since the camera is changing its position during the tracking, the background will also change (it will be static and then abruptly change). Different techniques will be used to address this problem. Another important aspect which must not be neglected is the fact that during the movement of the camera, if the estimation of the position of the subject is incorrect and the subject is “lost”, one must rapidly recover from this error in order to continue tracking the subject.
Applications: The main application of this research will be in surveillance systems, when a suspicious event occurs, one could continue the tracking for a longer period of time and with greater precision and detail so as to follow the person who has carried out the suspicious action.
Calendar: September 2004 - April 2006.
Last modification: 2007/09/28 by lbvailla


©2002-. Computer Vision and Systems Laboratory. All rights reserved