Hybrid Eye-Tracking System for Cursor Control: A Kalman Filter and Exponential Moving Average-Based Approach for Robust Face Tracking

Precise and efficient real-time eye-tracking is essential for hands-free cursor control, aiding accessibility and human-computer interaction. Traditional eye-tracking methods often suffer from latency, noise interference, and inaccurate tracking, limiting their practical usability. To address these challenges, we propose a Hybrid Eye-Tracking System (HETS) that integrates MediaPipe, Dlib, Kalman Filtering, and Exponential Moving Average (EMA) smoothing to enhance tracking accuracy and responsiveness. Our approach combines Kalman Filter-based motion prediction with adaptive noise reduction using EMA, reducing jitter and improving cursor stability. Furthermore, we introduce a multi-modal facial landmark detection strategy by fusing deep-learning-based MediaPipe Face Mesh with Dlib’s shape predictor, ensuring robust and adaptive tracking across diverse environments. The system operates in real-time with a lightweight architecture, making it suitable for assistive technology and hands-free computing applications. Experimental results demonstrate superior tracking accuracy and smoothness compared to conventional methods. For further details and access to the implementation, please visit our repository: https://github.com/balajisaba/hybrid-gazing