This project explores real-time gaze tracking using computer vision techniques. The idea was to build a simple game where the player controls actions using only their eye movements, no mouse or keyboard needed.
The system uses a webcam feed processed with OpenCV to detect the user's face and eyes. Roboflow Inference handles the gaze estimation model, determining where on the screen the user is looking. This gaze data is then mapped to game controls, allowing the player to interact with on-screen elements just by shifting their gaze.
The whole setup runs inside a Docker container for easy reproducibility. The project is open source — visit the GitHub repo eye-tracking-game to check it out!
Getting accurate gaze estimation in varying lighting conditions was the trickiest part. Small head movements can throw off the calibration, so I spent time tuning thresholds and smoothing the gaze signal to make the game playable.
Python, OpenCV, Roboflow Inference, Docker