Drowsiness Estimation of Drivers Using Echo State Networks

Abstract

Auto-estimation of drowsiness of a vehicle driver is an urgent problem, as more than 15% of traffic accidents are caused by careless driving in Japan. This paper considers drowsiness estimation based on the measurements of the heartbeat, respiration, and the center of pressure on the seat. As the learning model for the prediction, we employ an echo state network (ESN). An ESN is one of the recurrent neural networks (RNNs) that solves the problem of the high computational cost of ordinary RNNs by restricting the tunable parameters. As drowsiness at a time is dependent on the previous time, i.e., the system has dynamics, RNNs will be suitable models for drowsiness estimation. Furthermore, because of their low computational cost, ESNs are a promising candidate model for driver drowsiness estimation. The validity of the proposed method is tested by several persons' data.