Integrating OpenFace 2.0 Toolkit for Driver Attention Estimation in Challenging Accidental Scenarios

摘要

Gaze estimation is a typical approach to monitor the driver attention on the road scene. This indicator is of great importance in safe driving and in the design of the takeover control strategy for a Level 3 and Level 4 automation system. Nowadays, most of eye gaze tracking techniques are intrusive and costly, which limits their applicability over real vehicles. On the other hand, current databases used for gaze validation face the driver attention task focused on critical situations in simulation but they do not encounter actual accidents. This paper presents a low-cost and non-intrusive camera-based gaze mapping system integrating the open-source state-of-the art OpenFace 2.0 Toolkit [3] to visualise the driver attention simulation on prerecorded real traffic scenes through a heat map. The proposal has been validated by using the recent and challenging public dataset DADA2000 [9] which has 2000 video sequences with annotated driving scenarios based on real accidents. We compare our system with an expensive desktop-mounted eye-tracker, obtaining on par results and showing it is a good tool for driver attention monitoring able to be used in the design of take over systems and driving scenarios awareness systems for automated vehicles.