Human-Computer Interface systems are becoming a critical topic in many fields as it extends into more aspects of daily life. In particular that of face recognition has attracted much research interest that ranges of approaches to realize face detection in real time. Our study proposes an eye-blink detection model consisting of seven overall steps: processing of (a) video to frame, (b) pre-process, (c) face detection, (d) eye region localization, (e) eye landmark detection and status interpretation, (f) eye blink detection, and (g) eye blink classification. In the first step, frames are extracted from video sequences, recording the moment in time on each frame to perform subsequent actions. The proposed method adds an important module in the pre-processing phase, using KMF to improve the quality of frames. The Viola-Jones Model is used for face detection. In the eye region localization stage, the eye region is localized in detected faces which consists of two main steps (eye feature extraction and eye landmark detection). The CNN weights are refined with the help of an enhanced Seagull Optimization algorithm with enhanced exploration (SOEE), which is a further development of the classical seagull optimization algorithm (SOA) to ensure precise eye region localization. Following this, the optimized CNN framework generates the exact location of the eye region. Once the eye region is detected, the model identifies whether the eye is open or closed by calculating a value known as the Eye Aspect Ratio (EAR), which acts as a finger print of the eye and encodes its state. Finally, based on correlation coefficients computations, the detected eye-blinks are classified as a short or long eye-blink. A comparative evaluation confirms the efficacy of the proposed model.