Automatic detection of abnormal human events on train platforms

Abstract

Video surveillance systems that contain a large number of cameras makes the continuous monitoring of the video feeds nearly an impossible task. A transit or transportation authority usually deploys a video surveillance system to monitor and identify events in the system such as crowd behavior and crime. In this paper we present a method for automatically detecting people jumping or falling off a train platform. An experimental evaluation is described using a dataset that was recorded at a train station.