Description
Date depot: 1 janvier 1900
Titre: Video crowd detection and behavior analysis
Directeur de thèse:
Daniel RACOCEANU (ICM)
Domaine scientifique: Sciences et technologies de l'information et de la communication
Thématique CNRS : Non defini
Resumé:
The thesis works will cope with automatic detection and analysis of crowd in video sequences. This work addresses visual surveillance, which has been for ten years a major research area in computer vision.
Crowd image analysis is a key technology for public safety (e.g. for transportation networks, town centers, sport grounds) and for monitoring of transportation networks and public facilities.
The scientific challenge is to design and implement algorithms enabling to automatically obtain relevant information about the movements and the collective behavior of individuals and groups within a crowded scene observed by a single camera or by a network of cameras.
Behavior analysis in crowded scenes remains an open problem in computer vision due to the inherent complexity and vast diversity found in such scenes.
Two main technical problems have to be overcome which are:
• to identify flow patterns without tracking individual objects, which is both impractical and unnecessary in the context of dense crowds
• to understand changes in behavior when the scene context and crowd dynamics can vary over such a wide range.
Several methods based on optical flow have been presented in recent years to solve these problems. However, optical flow is only an estimation of instantaneous observed motion, and this “short-time” characteristic drastically limits the efficiency of analysis, which must be done on significantly longer time periods to efficiently represent spatial and temporal features of a flow that are useful for general applications.
Some promising new approaches propose to use concepts derived from fluid mechanics to overcome this issue. The thesis works will enable us to investigate this new domain.
Doctorant.e: Fagette Antoine