AcouBlinkDB: A Contactless Acoustic Eye-Blink Detection Dataset

We developed an acoustic eye blink detection system implemented on an HP ProBook 440 G5 laptop, utilizing its built-in speaker to emit inaudible continuous sound waves in the 18–22 kHz range. During each experiment, reflected acoustic signals are captured to identify eye blinks, and the resulting data are compiled into a repository comprising 670 labeled samples. The dataset includes 67 participants (35 males and 32 females), aged between 18 and 50 years and with heights ranging from 150–180 cm. Experiments were conducted across three environments—TV lounge, laboratory, and bedroom—to evaluate system performance under varied scenarios. Participants represented four distinct regions (South Asia, East Asia, Eastern Europe, and Africa), accounting for variability in eye size. The setup was tested over distances ranging from 50–75 cm, face orientations from 0° to 60°, and lighting conditions between 50 and 1450 lux. Eye blink speeds were categorized as slow (<10 brpm), normal (15–20 brpm), and fast (25–30 brpm). The system was also evaluated with various types of eyewear, including no glasses, regular eyesight glasses, polarized lenses, tinted lenses, and metallic frames. After each experiment, Fast Fourier Transform (FFT) values were extracted to differentiate blink and non-blink instances, which were labeled as 'B' (blink) and 'N' (non-blink), respectively.