Driver Visual Attention Estimation Using Head Pose and Eye Appearance Information

In autonomous, as well manually operated vehicles, monitoring the driver visual attention provides useful information about behavior, intent and vigilance level of driver. The gaze can be formulated in terms a probabilistic map representing region around which driver’s is focused. area estimated changes based on confidence estimation. This paper proposes framework convolutional neural networks (CNNs) that takes head pose eye appearance inputs, creates fusion model estimates 2D grid. contains upsampling layers to create estimations at multiple resolutions. trained using data collected from 59 subjects with continuous recordings where subject looks moving target parked car, glances set markers inside car while driving vehicle parked. Our superior performance than unimodal systems exclusively or information. It location lying within 75% an accuracy 92.54%.