Efficient machine learning approach for volunteer eye-blink detection in real-time using webcam

• DOI: 10.1016/j.eswa.2021.116073
• Publication Date: 2021-10-19T05:17:38Z
• Authors: Paulo Augusto de ...; Gabriel Vinícius ...; Felipe Ricardo do...; Ignacio Sánchez-G...; Hertz Wilton Cast...; Daniele Montenegr...; Danilo Alves Pint...; Ricardo Alexsandr...
• Keywords: Region of interest

The progressive diminishment of motor capacities due to Amyotrophic Lateral Sclerosis (ALS) causes a severe communication deficit. development Alternative Communication software aids ALS patients in overcoming issues and the detection signals plays big role this task. In paper, volunteer eye-blinking is proposed as human–computer interaction signal an intelligent Computer Vision detector was built for handling captured data real-time using generic webcam. eye-blink treated extension eye-state classification, base pipeline used delineated follows: face detection, alignment, region-of-interest (ROI) extraction, classification. Furthermore, complemented with auxiliary models: rotation compensator, ROIs evaluator, moving average filter. Two new datasets were created: Youtube Eye-state Classification (YEC) dataset, from AVSpeech dataset by extracting images; Autonomus Blink Dataset (ABD), completely result present work. YEC allowed training eye-classification task; ABD specifically idealized taking into consideration detection. models, Convolutional Neural Network (CNN) Support Vector Machine (SVM), trained performance evaluation experiments both models conducted across different databases: CeW, ZJU, Eyeblink, Talking Face (public datasets) ABD. impact evaluated CNN SVM compared classification Promising results obtained: 97.44% accuracy task on CeW 92.63% F1-Score dataset.