Towards a real-time situational awareness system for surveillance applications in unconstrained environments

Observing public spaces like car parks, airports, and train stations via video surveillance is an extremely tedious and error-prone activity for human operators. A generic real-time system is presented which closes the situational awareness loop from basic object detection, object tracking and conceptual situation recognition. The situation recognition is implemented as formal knowledge-based reasoning approach. In order to process information about objects in a scene a sophisticated multiperson tracker was integrated. The person tracking relies on local features. For person representation a generalized appearance codebook is used. The whole system is parallelized to gain real-time performance on ordinary hardware. The proposed system was evaluated on data reecting a prototypical surveillance scenario and promises practical results.