Integrated Passenger Flow Analysis and Street-Level Monitoring for Public Transport Management Using Deep Learning and IoT

Abstract

This paper proposes an integrated method for real-time passenger counting and street-level monitoring in public transportation vehicles. Unlike previous flow-based approaches, the proposed system leverages pre-trained deep convolutional neural networks, specifically the YOLO model, to analyze video streams captured both inside and outside the vehicle. The approach is implemented on GPU-enabled PCs installed onboard and connected to the cloud for transmitting data on vehicle occupancy and street-level observations. Internal video streams are processed to perform accurate passenger counting after doors are closed and flow analysis during stops, improving precision over flow-only methods. For external monitoring, additional cameras capture the environment around the vehicle to detect key objects such as potholes, garbage, and bicycles. These detections are geo-referenced using GPS data to generate informative maps highlighting urban street features. Furthermore, the passenger flow data is shown to support public transport network analysis, enabling insights into route usage and bottlenecks. The solution was validated in a large-scale deployment on 50 buses operating across 30 routes in the public transportation network of Bari, Italy, demonstrating high accuracy and practical applicability for urban mobility management.