Real-Time Object Recognition with Neural Networks in Public Transport – Determining the Utilization of Vehicles using Existing Camera Systems

Abstract

For security reasons, many local public transport vehicles now have cameras installed in their interiors. At present, these can and may only be used in Germany to investigate criminal offenses. Real-time object recognition offers the possibility of counting passengers with existing cameras without saving images or videos. This is not only important for revenue sharing, but can also provide information about bottlenecks when boarding and alighting, or be used to display empty areas in individual carriages of a train. This paper investigates whether object recognition is suitable for determining the utilization of public transport vehicles using individual images from the interior. Test images from the security cameras of a streetcar were analyzed and evaluated with a self-trained Faster R-CNN model. Accuracies of 70% were achieved in the detection of people and free seats.

Keywords: Passenger Counting, Artificial Neural Networks, Innovative Public Transport

DOI: 10.54941/ahfe1006222