A full-scale hardware solution for crowd evacuation via multiple cameras

Abstract

Crowd evacuation is thoroughly investigated in recent years. All efforts focus on improving safety standards of such a process. Past and latest life-threatening incidents related to evacuation procedures justify both the growing scientific interest as well as the interdisciplinary character of most research approaches. In this chapter, we describe the hardware implementation of a management system that aims at acting anticipatively against crowd congestion during evacuation. The system consists of two structural components. The first one relies on an elaborated form of the Viola et al. [55] detection and tracking algorithm, which incorporates both appearance and motion in real-time. Being supported by cameras, this algorithm realises the initialization process. In principal, it consists of simple sum-of-pixel filters that are boosted into a strong classifier. A linear combination of these filters properly set thresholds, thus succeeding detection. The second part consists of a Cellular Automata (CA) based route estimation model. Presumable congestion in front of exits during crowd egress, leads to the prompt activation of sound and optical signals that guide pedestrians towards alternative escaping points. The CA model, as well as the tracking algorithm are implemented by means of Field Programmable Gate Array (FPGA) logic. Hardware accelerates the response of the model by exploiting the distinct feature of parallelism that CA structures inherently possess. Furthermore, implementing the model on an FPGA device takes advantage of their natural parallelism, thus reaching significant speed-ups with respect to software simulation.The incorporation of the design as a fast processing module of an embedded system dedicated to surveillance is also advantageous in terms of compactness, portability and low cost.