Robust and Accurate Eye Blink Detection Using a 24 GHz CW Radar

Artikel i vetenskaplig tidskrift, 2025

A novel algorithm was proposed to address the challenges of eye blink detection, enhancing both robustness and accuracy. The key innovation of the algorithm includes a short-time iterative adaptive approach for acquiring higher-resolution time-frequency features, which facilitates subsequent processing; an effective denoising method based on probability estimation; and a dynamic adaptive segmentation method to minimize false alarms and miss alarms. The efficacy of the proposed detection algorithm was validated through experiment studies conducted in three different environments, each presenting varying levels of complexity. Our algorithm demonstrated superior overall performance compared to existing methods. With a low-cost 24 GHz CW radar, an average recall rate of 95.05%, a precision rate of 96.38%, and an F1 score of 95.68% were achieved at a distance of 50 cm.