Privacy-preserving crowd incident detection: a holistic experimental approach

Abstract

Detecting dangerous situations is crucial for emergency management. Surveillance systems detect dangerous situations by analyzing crowd dynamics. This paper presents a holistic video-based approach for privacy-preserving crowd density estimation. Our experimental approach leverages distributed, on-board pre-processing, allowing privacy as well as the use of low-power, low-throughput wireless communications to interconnect cameras. We developed a multicamera grid-based people counting algorithm which provides the density per cell for an overall view on the monitored area. This view comes from a merger of infrared and Kinect camera data. We describe our approach using a layered model for data aggregation and abstraction together with a workflow model for the involved software components, focusing on their functionality. The power of our approach is illustrated through the real-world experiment that we carried out at the Schönefeld airport in the city of Berlin.