Abstract by Brooks Butler
On the Informativity of Visual vs. Acoustic Data for the Detection of Violence in Crowd Control
This paper considers a new application of crowd control, namely, keeping the public safe during large scale demonstrations. This problem is difficult for a variety of reasons, including limited access to informative sensing and effective actuation mechanisms, as well as limited understanding of crowd psychology and dynamics. This paper takes a first step towards solving this problem by focusing on characterizing the informativity of sensor data. In particular, since many demonstrations occur at night or in situations where video access may be restricted, this paper explores the efficacy of acoustic vs. video data for identifying violence. Perhaps surprisingly, our results indicate that audio data is nearly equally effective at detecting violence as video data alone, and only slightly less effective than a combined signal.