Multi-Camera Crowd Monitoring: The SAFEST Approach
Résumé
This paper presents the current state of people counting approach created for the SAFEST project. A video based surveillance system for monitoring crowd behaviour is developed. The system detects dangerous situations by analysing the dynamics of the crowd density. Therefore we developed a grid-based people counting algorithm which provides density per cell for the global view on the monitored area. Since multiple cameras may observe same parts of the monitored area, the challenge is not only to count people seen by single cameras , but also to merge the views. Therefore we first detect people seen by each camera separately and then sum the results to a global representation. In order to avoid multiple counting of same objects, the output of cameras in the overlapped regions are weighted.
Origine : Fichiers produits par l'(les) auteur(s)
Loading...