TransitLK

BSc (Hons) Artificial Intelligence and Data Science | Data Science Group Project

Gowrisankar Sivakumar

Isira Sanjan Withana

Tariq Ramzeen

Vihanga Kukulevithana

TransitLK is a real time public transport monitoring and passenger engagement framework built for Sri Lanka’s 18000 buses, which currently operate with zero digital intelligence, no ETAs, no fleet visibility and no emergency response capabilities The system runs entirely on a single Android smartphone mounted on the bus using GPS and IMU sensors to power two separate pipelines. The GPS data streams to a cloud backend (Render) that runs an XGBoost based ETA prediction model, pushing arrival times to a passenger and admin apps in real time. The IMU data runs edge AI to detect crashes using multi sensor fusion, triggering automated SMS alerts to police, fire and hospital services in under 10 seconds Beyond sensor based crash detection, the system includes CV based crash detection, driver drowsiness monitoring, foul activity detection and passenger incident detection all running on edge without needing servers or cloud AI. A Firebase real time database ties everything together, feeding a passenger app, driver app and admin dashboard simultaneously The team validated the system with SLTB across a live 42 day pilot on route 256/341-1 using a 2020 Xiaomi Redmi 9, which is a 6 years old budget phone proving that low cost hardware can power production grade AI. Every deployed model outperformed published benchmarks. SLTB’s CEO and Deputy IT Manager reviewed the system and expressed interest in expanding it to school buses and urged us to contact more private bus operators as well. The project is the first of its kind to unify ETA prediction, crash detection and driver/passenger monitoring into a single smartphone based framework designed specifically for developing world transit condition”