Cognitive Score Prediction based on Transformer Encoder from Attentional Behavior of Elderly Drivers

Abstract

Although the number of fatalities from traffic accidents is decreasing in Japan, the percentage of fatal accidents involving elderly drivers is on the rise. In addition, the number of elderly drivers is increasing due to the aging of the population, and it is predicted that the percentage of accidents involving elderly drivers will continue to increase in the future. In order to reduce the number of serious traffic accidents, countermeasures for elderly drivers are necessary. To address this problem, this research aims to explore a new method for diagnosing the driving aptitude of elderly drivers. In this paper, we propose a model based on the LSTM-Transformer encoder named ELDANet (ELDerly Driving Assessment Network), which predicts cognitive scores or classifies cognitive categories based on the attentional behavior of elderly drivers, and demonstrate the effectiveness of the proposed model through experiments using videos of right-turn scenes.