Modeling Required Driver Attention Level Based On Environmental Risk Factors Using Deep Convolutional Neural Networks

Abstract

Understanding the level of environmental risk using vehicle-mounted camera traffic scenes is useful in advanced driver assistance systems (ADAS) to improve vehicle safety. We propose a fast, memory-efficient computer vision based environmental risk perception method using a weakly supervised convolutional neural network-based classifier. We use traffic scenes from Berkley deep drive dataset to evaluate the proposed method. Experimental results demonstrate that the proposed method correctly classifies required driver attention levels by considering multiple environmental risk factors. Further, we use class activation mapping to demonstrate that the proposed network is capable of identifying the underlying environmental risk factors.