optNet-50: An Optimized Residual Neural Network Architecture of Deep Learning for Driver's Distraction
Tahir Abbas, Syed Farooq Ali, Aadil Zia Khan, Irfan Kareem
Abstract
Over the last few decades, human facial recognition has gained significant popularity in areas ranging from surveillance, tracking, and access control to more recent developments in advertising, disease diagnosis, assistance to people requiring special needs and drivers' distraction detection. This study proposes a modified deep learning neural network architecture using Openpose library for a two-category problem of distraction detection. Openpose library, detects the human face and draws 43 points on face skeleton, which is given to the deep network along with the input images. The proposed approach attains an accuracy of 98% on publicly available `State Farm Distracted Driver Detection' dataset and outperforms existing state-of-the-art deep Residual Network architectures including ResNet-50 and ResNet-101.