Secure carpool system using vision verification and audio processing

Yee, Jia Hao (2026) Secure carpool system using vision verification and audio processing. Final Year Project, UTAR.

PDF Download (6Mb)

Abstract

University campuses are ideal for carpooling, yet adoption remains low due to safety, trust, and reliability concerns. This project introduces UniRide, a secure and scalable carpooling mobile application tailored for academic institutions like Universiti Tunku Abdul Rahman (UTAR). UniRide integrates multi-layered safety mechanisms for proactive protection, beginning with a pre-trip “Device Verification Guard” that prevents impersonation using ML Kit and a Zero-Trust Hybrid Cloud Architecture. This architecture utilises an encrypted Cloudflare Tunnel to a local Flask API, offloading ArcFace biometric processing to bypass cloud GPU costs while achieving sub-second verification latency. During trips, an edge-deployed YAMNet acoustic model operates directly on the smartphone to detect distress sounds locally. Anomalies immediately trigger a multi-modal validation pipeline using the Gemini 2.5 Flash/Lite API and IMU sensor data, confirming emergencies and dispatching automated alerts in an average end-to-end time of just 4.29 seconds. To ensure consistent service availability, UniRide features a semester-based scheduling system and a gamified Earn-On-Demand (EOD) economic model, allowing drivers to earn flexibly while keeping rides affordable for students. Developed with Flutter and Firebase, the modular application also includes QR-based authentication, geofencing, and dynamic routing with pass-by drop points to optimise match rates. Ultimately, by combining AI-driven safety and dynamic economic incentives, UniRide redefines campus carpooling as a trustworthy, sustainable mobility solution that actively supports SDG 11 (Sustainable Cities), SDG 13 (Climate Action), and SDG 16 (Peace and Justice).

Actions (login required)