Journal Search Engine
The Journal of The Korea Institute of Intelligent Transport Systems Vol.24 No.5 pp.159-181
실주행 데이터 기반의 LSTM 모델을 통한 실시간 자율주행차 이상상태 분류 연구
LSTM-Based Real-Time Classification of Abnormal Cases in Autonomous Vehicles Using Real-World Driving Data
Abstract
This paper proposes an LSTM-based model for real-time classification of abnormal cases in autonomous vehicles to enhance driving safety. We defined six representative abnormal cases by analyzing DMV disengagement reports and collected corresponding real-world driving data. We then constructed seven datasets through correlation and multicollinearity analysis, and trained the LSTM model using various combinations of hyper parameters. As a result, our model achieved a classification accuracy of 96.69% within 0.39 seconds, which aligns with the minimum human driver reaction time. This study is meaningful in that it demonstrates both high classification accuracy and fast inference time using real driving data. We expect that our approach will contribute to enhancing the safety of autonomous driving systems by integrating external data and expanding the range of abnormal case types.
