Driver Gaze Zone Dataset With Depth Data

摘要

Drivers' inattention and distraction detection are crucial features for driver monitoring and driver assistance systems (DAS) and a key point to avoid car crashes. Currently available datasets mainly track head pose and only a few were collected inside a car. Therefore, we present a novel public dataset designed to train algorithms that estimate driver gaze zone in real driving conditions. We provide labeled frame-by-frame images from 19 car points. Depth 3D images were also captured and they are aligned with infrared images. This new dataset is the largest one available for this purpose and the only one that provides 2D and 3D data aligned at pixel using a Intel^{? RealSense? camera. Finally, in order to establish a baseline, we present a gaze zone estimator built using a transfer learning method.}