Simulator of Vehicle-Mounted Camera Video to Verify Dynamic Saliency Maps for Predicting Drivers' Gaze Points
Rintarou Mizuno, Sorachi Nakazawa, Yohei Nakada · 2021 IEEE Symposium Series on Computational Intelligence (SSCI) · 2021
There has been remarkable development of driver assistance systems in recent times, and it is expected that these assistance systems will significantly contribute to reducing traffic accidents caused by human attention errors. To improve the functionality of such assistance systems, several studies have been conducted on computational models to compute dynamic saliency maps for predicting the gaze points of drivers. Against this background, this study introduces a prototype simulator that can produce vehicle-mounted camera videos in various driving situations using 3D computer graphics. This simulator easily enables us to validate the effectiveness of models for computing dynamic saliency maps in various driving situations, even dangerous driving situations that are difficult to capture by actual vehicle-mounted cameras. As an initial validation of this simulator, we reproduced vehicle-mounted camera videos in situations with other cars and pedestrians. Subsequently, we applied the models to compute dynamic saliency maps, which were proposed in our previous work to reproduce vehicle-mounted camera videos. The obtained results proved the usefulness of the proposed simulator. Improvements to our models have also been discussed.