iDAS — Driver Safety App

Distracted driving and missed traffic signals are responsible for a large share of road incidents in India. Most consumer ADAS systems are bundled with high-end cars; nothing for the average driver with a phone on the dashboard.

Build a cross-platform mobile app that uses the phone camera to detect traffic signs, signals, and lane markings in real time, then narrates warnings via text-to-speech. Has to work offline, has to be small enough to ship in a single APK / IPA.

Train a YOLO object detector on Indian road signs and traffic lights, quantise to TFLite, embed inside a Flutter app that exposes the camera frame to the model via platform channels.

• Dataset: scraped + hand-labelled ~6K Indian traffic-sign images.
• Trained YOLO weights, quantised to TFLite (~12MB final model).
• Lane detection ran as a separate classical CV pass (Hough transform on Sobel edges).
• Flutter UI with a single camera-preview screen and TTS warnings via the OS engine.
• Targets: Android, iOS, and Windows — same Dart code, three platform channel implementations.

Working prototype across all three platforms. Detects most common Indian sign categories with workable accuracy. Capstone project for the BE-MBA.

I would have started with a smaller class set and shipped narrower; I tried to detect everything and the real-time inference budget suffered.