UAV-based monitoring of pedestrian groups

In this work we discuss the problem how pedestrians can be monitored from an eye in the sky. Pedestrian monitoring is very important for large-scale surveillance of big events like festivals or demonstrations, and similarly for normal public places. At such settings it is not only important to know if an abnormal scenario is taking place in a pedestrian crowd in general, but also what specific type of scenario is taking place. Also, monitoring authorities benefit from an additional rating if the detected scenario is potentially dangerous to be able to react in an adequate manner. Therefore, we constructed a complex event detector (CED) to detect specific and potentially dangerous scenarios in pedestrian groups. For the development and evaluation of the CED we generated a novel dataset of moving pedestrian groups. The dataset was captured from an 8-rotor helicopter and consists of image sequences of 15 group behaviour scenarios in several complexity levels. We discuss the advantages of UAV imagery in comparison to aerial imagery captured from planes for the problem of pedestrian monitoring. Furthermore, we discuss the results of our CED for pedestrian groups which is able to detect scenarios which potentially occur in moving crowds.