Multi-EmoNet: A Novel Multi-Task Neural Network for Driver Emotion Recognition
Yaodong Cui, Yintao Ma, Wenbo Li, Ning Bian, Guofa Li, Dongpu Cao
Abstract
Driver’s emotion affects driving safety Hu et al. (2013), therefore monitoring driver’s emotion could benefit road safety. However, the complex illumination conditions in a vehicle cockpit significantly challenge the effectiveness of camera-based facial expression recognition (FER) systems. To solve this problem, we proposed Multi-EmoNet, a novel multi-task neural network, to classify human facial expression under illumination variations and to restore noisy images. Our experiments demonstrate these two tasks are complementary and together facilitate better network representation learning. Our approach obtains significantly better classification accuracy on images with illumination variation compared to the baseline networks. More importantly, the proposed multi-task network is a general architecture that can be applied to any noise involved image classification problem.