使用安全代理量度評估號誌化路口右轉側撞風險程度

Abstract

在諸多事故型態之中，右轉側撞為高嚴重程度之碰撞，因其碰撞之角度接近90度，若事故發生時通常發生於較脆弱面。而我國由於混流之複雜特性，直右共用車道常見於各地，若汽車或大型車輛右轉時，可能因車流阻擋或車體大小差異等因素導致未察覺後方直行機車，此類碰撞結果殊為嚴重。故本研究聚焦於號誌化路口右轉側撞相關之事故分析，以建立相關量化評估模式。我國常見之交通安全改善方法為事故法，亦即透過數年之事故肇事資料進行路口改善方案研擬，然蒐集耗時長、效益難以快速評估、依賴警方紀錄，且越多事故資料越容易進行安全評估之手段，有違交通安全之零願景考量。可行之先發法如交通安全技法（Traffic Conflict Technique, TCT）評估路口風險，可規避前述缺點與疑慮：其使用「安全代理量度（Surrogate measures of safety, SMoS）」藉由其他可觀察到之資料作為衡量安全依據。路側攝影即為交通安全技法中常見的潛在風險事件之蒐集工具，其鉅集之資料庫除有助於車流管理外，眾多文獻亦以之進行安全評估。然路側攝影易受陰影遮罩與側面投影產生之座標誤差干擾，故以無人機進行空拍獲取路口影像。本研究聚焦於號誌化路口「右轉側撞」與「同向擦撞」此類同向直行車與右轉車的碰撞型態。而蒐集右轉側撞事件資料時，發現此類衝突與同向擦撞可能為相同起迄方向，其衝突當下判定為潛在右轉側撞或同向擦撞，其區別可能僅係微小速度差或加減速之駕駛決策，故本研究提出「共同風險」之概念，以描述此類觀點。至此，將較警方提供之事故資料知曉更多資訊。後以三種子方法如考量共同風險之細部分類：將右轉側撞與同向擦撞分開配對，Pearson相關係數與Spearman排序係數可達0.7463與0.4696；以總和分類考量共同風險：將右轉側撞與同向擦撞之資料合併視為一類，Pearson相關係數與Spearman排序係數可達0.7882與0.6301；與不考量共同風險之純右轉側撞：亦即原始觀點不考慮同向擦撞，Pearson相關係數與Spearman排序係數篩選下為0.7889與0.6025。可見使用無人機進行安全代理量度資料蒐集判別之潛在風險事件，確實可以對應到事故數據。另外，本研究亦建立一模式整合子方法，以期僅使用觀察之事件數，預測可能之事故頻次，量化風險。以一例展現應用價值推估具改善前後影像資料之路口，在佈設直、右轉彎指向線分開之標線改善措施，其路口風險有所下降，推測其安全有所改善。

Among various collision types, right-hook turn collisions are recognized as highly severe due to their near-perpendicular impact angle, often involving the more vulnerable side of the vehicle. In Taiwan, the prevalence of shared lanes for through and right-turning vehicles—a result of mixed traffic conditions—significantly increases the risk of such collisions. When cars or heavy vehicles make right turns, they may fail to detect motorcycles traveling straight ahead due to traffic obstructions or differences in vehicle size, often resulting in severe outcomes. This study focuses on analyzing right-hook turn collisions at signalized intersections to establish a quantitative evaluation model. In Taiwan, the commonly adopted crash record-based approach to traffic safety improvements relies on multi-year crash data to propose intersection modifications. However, such methods are time-consuming, difficult to evaluate promptly, dependent on police records, and paradoxically require more accident data for effective analysis—contrary to the Vision Zero goal of eliminating traffic fatalities. As a proactive alternative, the Traffic Conflict Technique (TCT) provides a method for assessing intersection risk using Surrogate Measures of Safety (SMoS), which rely on observable non-crash events as indicators of safety performance. Roadside video surveillance is often used to collect potential conflict events for TCT applications and has been widely adopted in traffic flow management and safety assessment. However, it is subject to coordinate errors caused by shadowing and side-angle projections. To overcome these limitations, this study utilizes Unmanned Aerial Vehicles (UAVs) to capture overhead footage of intersections. This research concentrates on two types of same-direction crashes: right-hook turn collisions and same-direction sideswipe collisions. During data collection, it was observed that these events often share similar trajectories and directional paths. The distinction between the two often hinges on minor speed differences or the driver’s decisions regarding acceleration and deceleration. Thus, this study introduces the concept of “shared risk” to capture this overlap better, providing more nuanced insight than what is typically available from police crash reports. Three analytical sub-methods were employed: (1) a disaggregated shared-risk classification that separates right-hook and sideswipe conflicts, achieving a Pearson correlation of 0.7463 and Spearman rank correlation of 0.4696; (2) an aggregated shared-risk classification that treats both as one category, yielding the highest correlation values (Pearson = 0.7882; Spearman = 0.6301); and (3) a non-shared-risk classification that includes only right-hook collisions, which also produced relatively strong correlations (Pearson = 0.7889; Spearman = 0.6025). The results demonstrate that UAV-based collection of SMoS data can reliably correspond to actual crash records. A prediction model was further developed to integrate these sub-methods, enabling the estimation of crash frequencies based solely on observed conflict events for the purpose of quantifying intersection and approach-level risk. A case application at an intersection with UAV footage captured before and after the implementation of directional lane markings separating through and right-turn movements—showed a reduction in estimated risk, indicating improved safety outcomes after the intervention.