Video surveillance applied to a parking lot in the visible and infrared domain
Louis St-Laurent
Xavier Maldague (Supervisor)
Problem: This research project is devoted to a specific branch of computer vision: video security. The objective is to develop a video surveillance system which is autonomous and context-independent. This project involved all design aspects, including identification of the required specifications, component selection, system integration and development of low and high-level algorithms.
Approach: A context-independent surveillance involves an interpretation of the scene which is limited to the object level. No object recognition is performed and no information about the scene is used. This simplification allows the developed system to be flexible since one needs only to add an additional level of processing based on context-dependent information in order to adapt the system to a desired application. In order to orient the definition of the specifications, an outdoor parking lot surveillance application was chosen. The system can function both in indoor and outdoor environments.

The novelty of the designed vision system is that it integrates information obtained from both thermal and visible colour images. The use of a beamsplitter allows both fields of view to be registered, so that an image fusion can be performed on the pixel level with minimum software processing. The combination of data from the two frequency ranges allows a more efficient and reliable detection and tracking of the objects of interest. Moreover, the availability of thermal data significantly increases the possible applications of the system.

Calendar: May 2002 - July 2004
Last modification: 2007/11/08 by lstlaure


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