Survival Analysis to Collision Warning System Performance Evaluation

• Main Authors: Shih-Yuan Chao, 趙士淵
• Other Authors: 許添本
• Format: Others
• Language: zh-TW
• Published: 2019
• Online Access: http://ndltd.ncl.edu.tw/handle/6j69rn

碩士 === 國立臺灣大學 === 土木工程學研究所 === 107 === The Thesis focus on using survival analysis to the performance evaluation of the collision warning system. Survival analysis is a method analyzing the time as a dependent variable. The collision warning is to warn that the adjacent lane is coming a vehicle. The systems are set on the T-junction near Dormitory III and the T-junction near Science and Engineering Building II, which the common conflict types of former junction are left-turned crossing conflict and right-turned merging conflict and of latter junction are right angle conflict and left-turned merging conflict. Fly a drone to collect the trajectories of the vehicles intersect in an intersection, and manipulate the softwore “Tracker” to transform them into time-space data. The thesis uses Post Encroachment Time (PET) for left-turned crossing conflict and right angle conflict and uses Time to Collision (TTC) for merging conflict. Based on the result of Kaplan-Meier curve, the "Warning" curve is different from "No Warning" curve for right angle conflict and merging conflict. Then it uses Cox proportion hazard function for further discussion of the relating factors of the conflicts. The significant variables of left-turned crossing conflict are rear vehicle speed. Of right angle conflict are warning and if rear vehicle turns. The common significant variables of merging conflict are warning and the angle difference of two vehicles. In addition, other significant variables of right-turned merging conflict are if the straight car first arrived the intersection and the pedestrian interference.