- 學位論文
基於Yolo之無號誌路口行人辨識
Uncontrolled Intersection Pedestrian Recognition Base on Yolo
摘要
自民國97年起，台灣的車禍事件一直在逐年上升，尤其是行人路口事件成長幅度更是達到了171%，其中有許多特殊路口會造成交通意外的出現，例如無號誌路口與非常規十字路口等等。 本論文提出了一種基於Yolo(You Only Look Once)的方法，用於在無號誌路口進行行人與載具辨識，以用來判斷區域中行人和各式載具的關係，減少政府勘查所需的人力與時間成本，從而判斷該路口是否會存在潛在風險，或是需要在特殊時段有交通指揮或升級為有號誌路口。 針對實驗結果，此影像辨識演算法可應用於圖片、影片與即時串流，可分析當下區域的行人、汽車、機車與自行車，平均mAP達到85.36%，最高mAP來到92.5%。
並列摘要
Since the year 2008, traffic accidents in Taiwan have been steadily increasing year by year, with a particularly notable growth rate of 171% in pedestrian intersection incidents. Many of these accidents occur at specific intersections, such as uncontrolled intersection and unconventional crossroads. This paper proposes a YOLO (You Only Look Once)-based method for pedestrian and vehicle recognition at uncontrolled intersections. The aim is to assess the relationships between pedestrians and various types of vehicles in the area, thereby reducing the time costs required for government inspections. This approach helps determine whether the intersection poses potential risks, or if it necessitates traffic control during specific time periods or upgrading to a signalized intersection. This image recognition algorithm can be applied to images, videos, and real-time streams,enabling the identification of pedestrians, cars, motorcycles, and bicycles in the given area.The best mAP is 92.5% and the average mAP is 85.36%.