Computer Vision
DriverAware
Vision-based driver distraction detection that combines body, face and hand cues, built for my honours year project.
- Role
- Researcher & Developer
My honours year project: a vision-based, multi-cue framework for detecting driver distraction. Instead of relying on one signal, it watches three: the driver's actions (body camera), gaze (face camera) and hand position on the wheel (hands camera).
A model is trained for each cue on the Driver Monitoring Dataset (DMD) from 16-frame video clips, and the project investigates whether combining the cue models' decisions (decision-level fusion) improves detection. It compares progressively more advanced architectures under one shared preprocessing and training setup, so differences can be traced to the architecture itself.
Built with PyTorch, OpenCV and MediaPipe for the models, a FastAPI service, and an Angular client.