Video-understanding framework for automatic behavior recognition

We propose an activity-monitoring framework based on a platform called VSIP, enabling behavior recognition in different environments. To allow end-users to actively participate in the development of a new application, VSIP separates algorithms from a priori knowledge. To describe how VSIP works, we present a full description of a system developed with this platform for recognizing behaviors, involving either isolated individuals, groups of people, or crowds, in the context of visual monitoring of metro scenes, using multiple cameras. In this work, we also illustrate the capability of the framework to easily combine and tune various recognition methods dedicated to the visual analysis of specific situations (e.g., mono-/multiactors' activities, numerical/symbolic actions, or temporal scenarios). We also present other applications, using this framework, in the context of behavior recognition. VSIP has shown a good performance on human behavior recognition for different problems and configurations, being suitable to fulfill a large variety of requirements.