Development of a Video-based Real-Time Passenger Boarding and Alighting Counting System for buses to Reduce Fare Evasion under Varying Camera Angles and Operating Conditions

Abstract

The present work aims to develop a real-time video-based passenger counting system for urban bus routes to minimize fare evasion, a prevalent issue in several Asian developing nations. It utilizes two secondary datasets and two primary datasets from India, to capture diverse operating conditions. Initially, video-frames were extracted using OpenCV, and subsequently pre-processed to facilitate efficient training. Next, YOLOv8 model was pre-trained and integrated with DeepSORT algorithm to identify and track passengers in each frame. Afterwards, an area detection based method was used to determine the direction of travel, facilitating passenger boarding and alighting count. Results showed higher performance across different datasets, indicating the effectiveness of the developed system under diverse operating conditions. Such results provides an opportunity for transit authorities to implement the system across existing bus fleets (where older fleets impacting the camera stability, bus fleets with varying camera positions), during congested hours, with varying lighting conditions.