Recognition of Open and Closed Eye Using Residual Networks For Driver Drowsiness Detection: Original and CLAHE-enhanced Images

Residual Network (ResNet) architecture is implemented in camera-based drowsiness detection. ResNet50, ResNet101, ResNet152, ResNet50V2, ResNet101V2 and ResNet152V2 are performed to build the detection models. Moreover, this study also investigates the use of original images and Contrast Limited Adaptive Histogram Equalization (CLAHE) enhanced images as input. The original image processed using ResNet101V2 gives the best accuracy, precision, recall, F1-score and AUC values of 99.13%, 99.13%, 99.13%, 99.13% and 99.95%. This proposed method outperformed the previous studies. Furthermore, this study shows that the image enhancement technique is not required to improve detection performance.