Blink-Based Cognitive Load Detection in Noisy Environments

AbstractThis thesis presents a robust blink-based cognitive load detection system designed to operate reliably in noisy visual environments. Traditional eye-blink–based models often degrade when exposed to motion blur, low-light artifacts, camera noise, or inconsistent head pose. To address these challenges, the proposed method integrates blink-rate variability, eye-aspect ratio dynamics, and temporal smoothing techniques with noise-resistant feature extraction. A hybrid deep-learning architecture improves detection accuracy by filtering out background noise while preserving meaningful ocular micro-expressions. Experimental evaluation across high-noise and low-noise datasets demonstrates significant improvement in both accuracy and stability, enabling reliable cognitive load estimation in real-world conditions such as classrooms, driving scenarios, and digital workplaces. This work highlights the importance of robust visual preprocessing and offers a scalable foundation for next-generation human–computer interaction systems.