Study on Clustering Method of Driving Behavior Data Based on Variational Auto Encoder and Coupled-GP-HSMM

Abstract

In Japan, where the population is aging, traffic accidents caused by elderly drivers have become a social problem. The main cause of the accident is a decline in cognitive ability, and there is an urgent need to develop technology for early detection of decline in the ability. The authors have been developing a technology to evaluate the cognitive ability of a driver from the driving behavior data. In the driving behavior data, there are some scenes where the cognitive ability can be evaluated and some scenes where the cognitive ability cannot be evaluated. Therefore, in this paper, we propose a clustering method for driving behavior data. Here, it is considered that the cognitive ability is the accuracy of the operation for the surrounding situation, and it is useful to cluster the relationship pattern between the situation and the operation for the driving behavior data as a group. In this method, Coupled-GP-HSMM and Convolutional Variational Auto Encoder are applied as a time series clustering method. This method realizes clustering of the relationship pattern between the time-series data representing the situation and the time-series data representing the operation.