無人機空拍應用於交通安全分析──號誌時制設計與風險駕駛行為

A Traffic Safety Analysis with UAV-based Trajectory Data—Signal Timing Design and Risky Driving Behaviors

本研究運用無人機空拍影像進行交通衝突與行為分析，透過人工智慧深度學習技術由影像中產出車流軌跡資料，計算駕駛人在路上可能產生事故風險之駕駛行為，並透過號誌燈側拍及對時，記錄駕駛行為與號誌燈號之間的關係，分析號誌設計時相配置對路口交通安全之影響。本研究以具風險之駕駛行為、違規行為、交通衝突三大類別風險駕駛行為之架構，提出一系列的相關指標進行量化，並透過易肇事路口案例分析，發現號誌燈號時相、黃燈及全紅秒數、標線及車道配置對車流行為安全性有關鍵影響，案例中亦發現號誌時相早開可能會對駕駛人帶來額外的交通安全風險，在制定合適方案時應慎重考慮其車流條件。研究結果可作為制定及評估交通工程改善方案之參考依據。

This study proposes the use of Unmanned Aerial Vehicle (UAV) aerial imagery for traffic conflict analysis. By applying artificial intelligence and deep learning techniques, vehicle trajectory data are extracted from the videos to evaluate driving behaviors that may lead to traffic crash risks. Additionally, by employing side-shot imagery of traffic signals and applying a video synchronization technique, the relationship between driving behaviors and traffic signal is analyzed, thereby assessing the impact of traffic signal design on intersection safety. A framework of three“Risky Driving Behavior”categories, namely risky driving behaviors, violations, and traffic conflicts, is established, and a series of related indicators are introduced for quantitative analysis. Through a case study of crash-prone intersections, it is found that signal phase design, yellow and all-red intervals, as well as road markings and lane configurations, play critical roles in traffic flow safety. Findings also reveal that early green signals may impose additional risks to road users, highlighting the importance of carefully considering traffic flow conditions when formulating appropriate signal timing plans. The research outcomes can serve as a reference for the formulation and evaluation of traffic engineering improvement measures.