Citation

Kato M, Emura K, Watanabe E. Trans. Soc. Automot. Eng. Jpn. 2022; 53(6): 1108-1113.

Copyright

(Copyright © 2022, Society of Automotive Engineers of Japan)

DOI

10.11351/jsaeronbun.53.1108

PMID

unavailable

Abstract

In this paper, in order to clarify the relationship between human prediction characteristics based on prediction coding theory and traffic near miss incidents, analysis for the front video of drive recorders recorded traffic near miss incidents was conducted using deep learning model which simulates human vision, and two hypotheses were proposed. Using the prediction error indicator based on the hypothesis, it was confirmed that 30 out of 60 near miss video can explained by the hypothesis. It was indicated that the change of the prediction error effects the attention of the unconscious and may lead the traffic near miss incidents.

Language: ja

Keywords

deep neural networks; drive recorder; human engineering; human error; machine learning; near-miss analysis; predictive coding; recognition; visual system