Fine-Tuning GMM and Total Pixel-Based Drowsiness Detection: A Strategy for Detection Open and Closed Eye

Abstract

Fatigue driving represents a substantial and often unrecognized risk in traffic accidents. A technique that may be employed involves the detection of open and closed eyes. The research on open and closed eye identification use approaches based on haar cascade and complete pixel analysis. Our proposed method employs an adaptive thresholding technique is implemented right before total pixel process. The processing steps involve the application of haar cascade, adaptive thresholding, fine-tuning of Gaussian Mixture Models (GMM), and the calculation of the total pixel count in the image that is utilized to identify the state of the eye using thresholding. The results from Fine-Tuning GMM thresholding for the left and right eyes are as follows: MSE values of 7.02 and 7.96, and PSNR values of 39.24 and 39.21, respectively. The results derived from fine-tuning are comparable to those obtained using Otsu's method.