Low-level segmentation and tracking of people with a network of independent cameras
Nicolas Martel-Brisson
André Zaccarin (Supervisor)
Problem: The detection and tracking of people in interior and exterior environments have been the subject of research and numerous publications for a multitude of applications. The complexity of the task depends on several factors including the type of background (fixed, mobile, evolving…), the quality of segmentation desired (with or without shadow, only the face…) and the complexity of the subject to be tracked (contrasting with the background, immobile, several people in the field of view...).
Motivation: The automation of the video surveillance domain is intrinsically linked to the capacity to detect subjects present in the image. Within the MONNET project (MONNET = Monitoring of Extended Premises: Tracking Pedestrians Using a Network of Loosely Coupled Cameras), we wish to segment the people in the background, follow them from one field of view to another and analyze their movements using a network of cameras where each unit of acquisition and treatment is independent. This independence enables the minimum of information to be transferred between the treatment nodes. The segmentation of people is the initial step in a multitude of movement and tracking algorithms and analyses.
Approach: In order to adequately segment silhouettes present in the camera’s field of view, the background will be modeled with a sum of Gaussian distributions. This statistic model will be combined with a basis movement estimator based on block matching. This estimator will allow a spatial treatment of the segmentation, which is lacking from the statistic model. The detected subjects will then be tracked throughout their presence in the camera’s field of view by following the movement vectors and recognizing them in front of other cameras by studying their statistic distributions.
Challenges: The main challenge in this research is to adequately combine several algorithms of modeling and tracking in a cooperative manner.
Applications: This project has numerous applications in the area of surveillance and artificial vision.
Calendar: January 2003 - January 2005
Last modification: 2007/11/24 by nmartel


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