Citation

Tanaka Y, Bando T. Trans. Soc. Automot. Eng. Jpn. 2013; 44(2): 685-690.

Copyright

(Copyright © 2013, Society of Automotive Engineers of Japan)

DOI

10.11351/jsaeronbun.44.685

PMID

unavailable

Abstract

Number of traffic accidents keeps decreasing, however, human error is still key factor of traffic accidents. Early detection of anomaly driver state is expected to contribute for further reduction of traffic accidents. In this study, we propose novel anomaly state detection method based on vehicle data observed in CAN. Acceleration and relevance velocity to a leading vehicle were employed and relationship between them is modeled using one class support vector machine (OCSVM). OCSVM enables anomaly detection without gathering anomaly behavior data. Effectiveness of our anomaly detection method is evaluated using driving data includes both of highway and ordinary road, and anomaly level estimated by our method is correlated with subjective human evaluation value.

Language: ja

Keywords

anomaly detection; driver model; human engineering; human error