An Industrial Visual Surveillance Framework Based on a Pre-Configured Behavior Repertoire: A Practical Approach

Abstract

We provide a practical industrial visual surveillance framework based on the notion of visual trap points. Instead of using the whole machinery of computer vision in order to verify correct workflow execution we re-factor the behavior training module to a pre-configured pool of allowed behaviors. We exploit humans' ability to distinguish tasks and allow for an automated surveillance system to accomplish the surveillance phase. Computer vision methods are used only for the object detection and recognition, and for this reason are re-positioned to the lower levels of an architecture for surveillance systems.