Effective Pedestrian Detection and Counting at Bus Stops

Abstract

This paper presents two approaches for pedestrian detection and counting at bus stops, whose purpose is to provide accurate demand estimates for an automated fleet scheduling and dispatching system. The approaches are general and can also be applied to people counting in public buildings or pedestrian detection from robotic platforms with minor modifications to account for background motion. The first approach employs a background subtraction scheme combined with the Viola and Jones detector, while the second one uses the foreground pixels count and a linear model involving people density estimates to calculate the total number of people. It is shown that the second approach is more accurate and reliable despite its simplicity, because it is not subject to the detection errors arising in the face or body parts detectors of the first approach. Both approaches are tested using image sequences from a real bus stop.