Invited Speaker---Prof. Toshio Ito


Development of Machinery and Control Systems, Shibaura Institute of Technology, Japan

Biography: Dr. Toshio Ito was awarded his B.A. in Engineering from Kobe University in 1982, and joined Daihatsu Motor Co., Ltd. He earned a Ph.D. in Engineering from the Graduate School of Engineering, Kobe University in 1995. He had been working on the research and development for advanced driver assistance systems, and commercialized pre-crash safety systems. He retired the Daihatsu in 2013, and was appointed as a professor in the Shibaura Institute of Technology. He specialized in the study of HMI for advanced driving assistance systems. His primary research interests are estimating the driver state using CAN data analysis from the point of view machine learning. He is presently conducting project on developing an automated driving mini EV, with a focus on analyzing the driver state by a heart rate sensor.

Speech Title: Predicting traffic congestion using driver behavior
Abstract: Traffic congestion is one of the largest traffic problems and one solution is a prediction of its generation. The traffic information provision system was developed in response to this issue. This system broadcasts traffic congestion information obtained from fixed monitoring points to vehicles in motion in real time to allow them to avoid traffic congestion, however, as it requires infrastructure investment, it has only been installed on roads with high traffic volumes such as major trunk roads and highways. It does not cover all roads. Traffic congestion comes from three phases order: the free travel phase, the meta-stability phase, and the traffic congestion phase]. Therefore, it can be considered that if the meta-stability phase can be detected, forecasting traffic congestion becomes possible. This paper proposes a driver model that forecasts traffic congestion based on big data analysis, that is, changes in driving behaviour and that does not rely on traffic flow monitoring infrastructure. As a result of evaluation in driving simulators, it was understood that the distribution of steering, throttle and speed input frequency changes based on changes in the travel phase. It is possible to distinguish these changes using support vector machines, and it is possible to make this into a driver model that predicts traffic congestion. This method is the first propose that uses only CAN (Controller Area Network) data and needs no additional sensors to detect driving environments or any infrastructures.