Real-Time Intentional Eye Blink Detection Using Rhombus Identification

Abstract

Eye blinking is an indicator for various applications such as face recognition, drowsiness detection, phone unlocking, etc. Existing work on blink recognition typically considers the normal expressions that the two eyes are unintentionally opened and closed simultaneously. There is no literature on detecting intentional eye blink with simultaneous one eye open and one eye closed. Such intentional eye blinks can serve as a customized signal for various applications, such as activating an intelligent security system by a member in danger. Further, the existing work uses the outline of an eye (called the Eye Aspect Ratio (EAR) method) to judge whether to blink or not. However, the sizes of people’s eyes could be different, and this method is thus prone to misjudgment. To remedy these loopholes, we propose the following two methods: (1) An intentional eye blinking mechanism with simultaneous one eye open and one eye closed as a signal to activate an intelligent security system; (2) A novel, fast, and accurate Rhombus Identification Method (RIM), considering the eye contour and normalized area, which can accurately identify expressions with eyes open and closed simultaneously, as a signal to activate security alarms. Compared with the EAR method, our RIM has significantly higher accuracy rates under different situations. Based on machine learning, we incorporate our RIM into an intelligent security system and justify the practicability of our RIM.

This Projet is Partially Sponsored by the MediaTek Foundation.