Time Series–Based Detection on Tailgating Fare Evasions Using Human Pose Estimation

Abstract

Passengers pass ticket barrier gates without paying in metro stations all over the world. This kind of behavior is called fare evasion, and it is troublesome and costly to prevent. As a typical type of fare evasion, tailgating refers to following a fare-paying passenger through the gate. It can be dangerous because the passenger risks being injured by the barrier gate. To detect tailgating fare evasions automatically, the existing surveillance cameras in stations can be utilized to provide a visual-based method at a low cost and efficiently. However, occlusion by crowds during rush hours can lower the accuracy of regular recognition methods based on convolutional neural networks. Moreover, the behavior of a tailgater can be similar to that of other fare-paying passengers if the positional relationship is not taken into account. Thus, we propose a tailgating recognition method that uses videos as input. First, the estimated human pose data in each frame is obtained, of which incomplete skeletons are retained. Second, the multiple persons appearing in adjacent frames are matched, after which a sequence of skeleton data is generated for each pedestrian. Third, a time series of the positional relationship between passengers and the ticket barrier gate is extracted and the passing interval of passengers is defined as the indicator for detecting tailgating. Our experiments showed that tailgaters could be distinguished effectively from fare-paying passengers, and the time series can cope with missing joints caused by occlusion or misidentification in a few frames.