Takeshi Matsumura

Driving skill analysis using machine learning The full curve and curve segmented cases

Analysis of driving skill/driver state can be used in building driver support and infotainment systems that can be adapted to individual needs of a driver. In this paper we present a machine learning approach to analyzing driving maneuver skills of a driver that covers both longitudinal and lateral controls. The concept is to learn a driver model from sensor data that are related to driving environment, driving behavior and vehicle response. Once the model is built, driving skills of an unknown run can be classified automatically. In this paper, we demonstrate the feasibility of the proposed method for driving skill analysis based on a driving simulator experiment in a curve driving scene for both the full curve and curve segmented cases.

Dominant Driving Operations in Curve Sections Differentiating Skilled and Unskilled Drivers

Our objective is to develop a new driving assist system that can help low-skilled drivers improve their driving skill. In this paper, we describe a statistical method we have developed to extract distinctions between high- and low-skilled drivers. There are three key contributions. The first is the introduction of wavelet transform to analyze the frequency character of driver operations. The second is a feature extraction technology based on AdaBoost, which selects a small number of critical operation features between high- and low-skilled drivers. The third is a simple definition for high- and low-skilled drivers. We performed a series of experiments using a driving simulator on a specially designed course including several curves and then used the proposed method to extract driving operation features showing the difference between the two groups.