The state of alertness plays a pivotal role in safety-critical operations, with reduced alertness posing risks to public health and safety. This paper focuses on measuring human alertness levels through non-intrusive, automatic, and cost-efficient means—specifically, utilizing distinctive facial features, such as saccadic eye movements. While current methods often rely on expensive proprietary eye trackers, our work explores the challenges and efficiency of employing a camera module for saccadic eye movement analysis. We trained a custom landmark detector to locate eye features, addressing localization challenges caused by low resolution and illumination variations. To enhance accuracy, we propose the use of a median filter to mitigate outlier effects. Additionally, we model an Extended Kalman Filter (EKF) to track eye coordinates, enabling the extraction of saccades and their velocity profiles. By mapping these profiles to Eigenspace, we successfully distinguish between alert and non-alert states. Experimental results demonstrate a classification efficiency exceeding 95% for the classification of states from saccadic velocity profiles.