AdapBlinker: Robust adaptive median filter approach to detect subtle eye blinks

Nowadays, eye blink detection is gaining significant attention in human-computer interaction systems. Users are increasingly favoring interactions with their phones and computers through non-manual methods, underscoring the constraints of conventional touch interfaces. Wearable technology, such as electrooculography (EOG)-based approaches and infrared sensors (IR), can accurately detect eye blinks; nevertheless, they can be inconvenient after prolonged use. Despite this, the drawbacks of camera-based eye blink recognition techniques are blind spots and the lighting effect. Thus, this study proposes an acoustic signal-based eye blink detection system to overcome these constraints. Acoustic signals can perform fine-grained detection within localized range due to their high attenuation in the air medium. The main benefit of acoustic sensing over conventional methods is that it senses signals directly, so the user does not need to wear any sensors. The prevalence of speakers and microphones in devices is another advancement that supports acoustic sensing. In this research, we present AdapBlinker, which employs the HP ProBook 440 G5 laptop to generate acoustic signals, retrieve data, process acquired signals, and plot Fast Fourier transform (FFT) to extract eye blink signals. AdapBlinker uses an adaptive median filter that adapts to surroundings, eliminates intrusions, and detects subtle blinks. We tested AdapBlinker with thirty-four participants across three settings for five months, achieving an average eye blink detection accuracy of 97.2%.