Automobile Driver Skills Assessment Focusing on Modifications of Operation and Causes of Modifications

Abstract

A system that evaluates driving skills based on driving data is important for realizing a society one drives safely. In this study, we defined the driving that fellow passengers feel safe as “driving the automobile as the driver intends and with little modification of the driving operation.” Then, we devised an algorithm for detecting “modification of the driving operation” based on the steering angle and the longitudinal acceleration of the driving time-series data. Furthermore, in order to identify the cause of the modification and feedback the cause to the driver, we constructed a classifier that predicts the cause of the modification at the detected point. As a result of learning and inference, we confirmed that the classifier predicted the correct cause with 0.199 F-measure in the same course as learning data.