Architecture for and evaluation of situational analysis in the real world

Automatic situational analysis in video surveillance enables smart and exhaustive observation of what is happening in a scene. In real world applications there arise special challenges concerning imperfect input data and real-time processing requirements. In this paper a hierarchical and distributed architecture tackling situation analysis in the real world is introduced, described, and exemplarily evaluated. The architecture introduced is hierarchically and modularly structured. The low-level data acquisition and processing is encapsulated in self-contained building blocks. These modules include video acquisition, sensor feedback and control, person detection and tracking, and vehicle identification. High level situation analysis is provided with a knowledge based approach using Fuzzy Metric Temporal Logic and Situation Graph Trees. The communication between the modules is realized with a persistence layer serving simultaneously as the knowledge base for the inference engine. Finally, the overall performance of recognizing situations is evaluated.