UTAR bus tracker: using AIOT for real-time tracking and occupancy estimation

Ng, Jia Shan (2026) UTAR bus tracker: using AIOT for real-time tracking and occupancy estimation. Final Year Project, UTAR.

PDF Download (6Mb)

Abstract

The lack of real-time information in university shuttle bus services creates significant difficulties for students, including uncertainty about bus arrival times, inability to determine seat availability, and absence of emergency communication. Existing applications such as Moovit, Google Maps, RapidKL Pulse, and Justnaik do not serve UTAR students as they either do not track UTAR buses, lack occupancy information, require continuous internet connectivity, or lack emergency reporting. This project applies Artificial Intelligence of Things (AIoT) to address these limitations through a real-time bus tracking and occupancy estimation system. The system consists of a Raspberry Pi 4 with a Neo-6M GPS module and 5MP camera running the YOLOv8n object detection model for passenger counting. The hardware performs edge-based passenger counting at 640×480 resolution with a confidence threshold of 0.3, processing one frame every 10 seconds while preserving passenger privacy by never uploading raw images to the cloud. The system transmits GPS coordinates and passenger counts to Firebase at 10-second intervals, while a Flutter mobile application with role-based interfaces for students, drivers, and administrators caches data locally using Hive for offline-first functionality. The system was evaluated through GPS accuracy testing (30 samples along the UTAR–Westlake route), passenger counting accuracy against manual counts (50 frames across 5 occupancy levels), Firebase data transmission testing, and offline mode testing across seven scenarios. GPS accuracy achieved an average error of 0.40 metres (min: 0.28m, max: 0.52m). Passenger counting accuracy reached 94.1% (MAE: 1.18 persons, RMSE: 1.52 persons), exceeding the 90% target. All offline mode test scenarios passed successfully. End-to-end system latency from camera capture to mobile display averaged 1.15 seconds. The system successfully demonstrates that AIoT technologies can effectively improve university shuttle bus services through real-time tracking, occupancy estimation, emergency reporting, offline functionality, and role-based access.

Actions (login required)