Appearance-based Driver's Gaze Mapping Using a Dash Camera
Ulziibayar Sonom-Ochir, Stephen Karungaru, Kenji Terada, Altangerel Ayush · 2022 Joint 12th International Conference on Soft Computing and Intelligent Systems and 23rd International Symposium on Advanced Intelligent Systems (SCIS&ISIS) · 2022
Recent gaze mapping studies indicated that consideration of both head position and eye gaze can benefit performance. Pushing this idea further, we propose an appearance-based method that, uses a combination of head position and eye gaze features. It is a low-cost, non-intrusive, and lightweight gaze mapping method that is not difficult to use or fatigues the driver. We proposed different strategies for the gaze estimation strategy using head position and eye gaze features. Within this proposal, we investigated combining the features one by one and with the other features in binary, triple, and quadruple to determine how they affect the estimation of gaze mapping. In addition to features, we also investigated how different camera positions affect gaze estimation. From these strategies, the strategy using OpenFace with SVM classifier (using gaze angle, head position_R features, head rotation_R features, and eye position WO-Z features) using camera position 2 outperformed all other methods and achieved a SCER rate of 85.64%, and LCER rate of 98.69% on the open dataset CAVEDB.