Low-cost in-bus people counting system for the reordering of urban passenger traffic based on artificial vision and neural networks
Gustavo Recalde, Derlis O. Gregor, David Britez, Mario Arzamendia · 2021 IEEE CHILEAN Conference on Electrical, Electronics Engineering, Information and Communication Technologies (CHILECON) · 2021
Due to the high rate of passenger traffic in public transport in Paraguay, it has become indispensable for public transport companies to implement specific methods for the organization and management of the service. Transportation companies have begun to take into account passenger counting, which could optimize the deployment of the vehicle fleet, make a more appropriate budget distribution among the different companies, and improve the service in general. This work aims to provide a low-cost analysis system for bus passenger traffic, geo-referencing the GPS positions where people get on and off the bus, also adding data based on parameters such as weather, temperature and humidity, among others. The counting system consists of a Jetson Nano running a MobileNet V2 SSD network for person detection and a tracking algorithm in conjunction with a CSI camera placed in a zenithal position to obtain the images. Data acquisition and network training were performed entirely on the Jetson Nano. The results show that the developed system can process video at a resolution of 640x480 pixels at 30 fps, and can count passengers in different lighting conditions with an accuracy of 94%.