運用行車影像事故資料分析事故發生過程-以國道客運為例

Abstract

由於警方事故調查報告未具備充分事故前資訊(pre-crash)，以致於事故成因的分析，以及改善方案的發展有其侷限性。儘管事故前資訊可透過影像畫面以及數位行車紀錄器蒐集，但資料有事件發生的先後順序，隨時間動態變化的特性；目前缺乏相關分析方法，使得研究人員無法有效找出事故肇因，以及量化事故的生成條件。有鑒於此，本研究提出一套系統化的資料編碼分析架構，以進行事故肇因的辨識與歸納；利用大型車輛右轉彎側撞事故為分析案例。本研究架構包含以下主要步驟：（1）建立行車影像資料為主的事故發生過程編碼架構，含駕駛人行為與行車動向情境，以將事故前資訊納入分析；（2）建立病例對照分析步驟(case-control study)，以驗證及歸納事故肇因；（3）建立事故發生過程之數學模式，更進一步瞭解分析事故發生的生成條件。本研究以實際發生事故的影像資料進行資料編碼，並透過病例對照研究辨識出右轉彎側撞事故的主要肇因，為駕駛人的注意力分配過程缺失。藉由數學模式的建立，本研究發現主要影響參數為大型車與其右方車輛的相對速度差，以及大型車駕駛人觀看右側照後鏡，至其實際轉動方向盤之間的時間長度。建議駕駛人應該避免注意力過度停留在右前側的位置，根據影像內容的結果，車身右後側是容易有潛在危險，並發生危害的重點區域。

Abstract Without pre-crash data has its limitation, when using police accident report to analyze the causes of the crash, as well as to develop the effective countermeasures. Despite pre-crash information can be collected through the video screen and digital driving record, a sequence of events is specific characteristic of data. Lack of available and useful methods to study the dynamic characteristics change over time, it’s very difficult to identify crash causality, and to quantize crash condition. This goal of this research is the development of basic coding scheme, to define, screen and identify crash systematically. Take the side-impact crash of large vehicle for example. The major steps are: (1) to establish basic coding scheme by screening pre-crash information of the video data, including driver behavior and driving situation；(2) to identify crash causality by using case-control study；(3) to further understand the formation conditions by building a safety frontier model. The results of this study indicate that the potential pitfalls of such vision shift are illustrated. The driver would be easy to expose at risk, if there was a period between glancing on the right side mirror and actually rotating the steering wheel. The ability to monitor rear right position which is potential sources of hazards is essential when right turn maneuver. The driver must prioritize locations in the visual scene according to their importance and frequently monitor the most likely hazard spots.