논문 상세보기
pp.33-48
This study proposes a data-driven framework for analyzing freeway driving behavior using multiple real-world trajectory datasets, and applies it consistently to mainline and ramp sections. The four large-scale datasets—namely highD, exiD, NGSIM I-80, and NGSIM US- 101—were processed through a unified preprocessing pipeline that converted all variables to International System Units(SI), resampled trajectories to 10 Hz, applied Savitzky-Golay smoothing to speed, and removed physically implausible and statistical outliers based on joint physical-statistical criteria. For each vehicle, 24 summary features were constructed from six longitudinal indicators–speed, acceleration, deceleration, time headway (THW), distance headway (DHW), and time-to-collision (TTC)–using their minimum, maximum, mean, and standard deviation. Indicator distributions by road type were compared using relative frequency histograms with common binning; then, principal component analysis (PCA) and K-means clustering were applied independently to each dataset. The leading principal components revealed interpretable axes related to longitudinal driving intensity (speed and acceleration level), safety margin (THW/DHW/TTC), and onramp sections; responsiveness was characterized by acceleration-deceleration variability, as observed within the analyzed datasets. Cluster interpretation yielded four relative driving behavior categories–aggressive, responsive, stable, and defensive–defined within each dataset based on indicator levels and variability rather than absolute thresholds.
1. 서론
2. 선행 연구 검토
3. 방법론
3.1. 데이터 수집
3.2. 데이터 전처리
3.3. 데이터 분석
4. 분석 결과
4.1. 지표별 분포 분석 결과
4.2. 클러스터링 분석 결과
4.3. 도로유형별 운전행태 분석
5. 결론
감사의 글
REFERENCES
