Institut Français des Sciences et Technologies des Transports, de l'Aménagement et des Réseaux

Type de document

CHAPITRE D'OUVRAGE (CO)

Langue

anglais

Auteur

BRUYELLE ; KHOUDOUR ; AUBERT ; LECLERCQ ; FLANCQUART

Résumé / Abstract

To improve security and passenger flow management, public transport camera networks gets bigger and bigger. However, operators are only able to monitor a few cameras at any given time. This fact makes such a system more and more inefficient as the number of cameras grows. It is therefore highly desirable to automate the monitoring tasks through computer vision. Although considerable work has been carried out in the field of image processing to perform detect events and measure flows for public transport operators, effective implementation of these tasks in a real-world network has not been fully realised. The working habits of the operators, directly derived from the screen wall concept, tend to favour the transmission of baseband video signals to a centralised processing computer. This approach is not well suited to computer-based image processing. We propose a different approach, consisting of small, localised computers dedicated to local image processing of one or two cameras only. These processors are interconnected through a computer network. The bandwidth of such network is sufficient even for a large number of cameras, since it only carries the small amount of useful data extracted from image processing and only when required (when an incident occurs, or at specific time intervals for traffic monitoring tasks). This chapter describes the implementation of this system, as well as four computer vision applications that have been implemented on it. Two of them are linked to security purposes and deal with intrusions and abnormal stationarities. The two others are related to passenger flow measurements and perform counting of passengers and queue length measurement. The algorithmic aspects of these processes are described in this chapter. They have all been intensively tested in real life conditions within several types of network such as an airport and underground railways. Trial results show the efficiency of the tested solutions.

Editeur

Inspec/IEE