基於YOLO v4影像辨識技術之智慧型機車安全距離警示系統

因台灣土地狹窄，人口密度高，具備靈活及高機動性的機車成為一般人最常使用的移動交通工具，近年來雖然有逐漸在倡導「防衛性駕駛」但交通事故仍無明顯的低減，在交通事故常造成人員傷亡，而究其交通事故的原因，「未依規定讓車」及「轉彎不當」佔比最高，而機車駕駛人常常與大型車爭道，會因大型車視線盲區不易察覺週遭車況而容易發生交通意外事故，近年來人工智慧也逐漸應用在我們生活上，本研究利用偵測準確度高且運算速度快的YOLO v4深度學習技術並配合影像辨識方法偵測出於街道中移動的大型車輛，藉由標記影像特徵且利用深度神經網路進行模型訓練模型，對於想要偵測的目標物做影像辨識。同時偵測與目標物的相對距離，當車輛進入到安全距離時，會透過警示來提醒機車駕駛人事前做出防範的措施，以減少機車的交通意外事故發生。實驗結果證明，本研究使用AI影像辨視方法進行大型車輛偵測其偵測的準確率結果為94%，是一個相當不錯的結果。

關鍵字

深度學習；機器學習；類神經網路； YOLO ；影像辨識

並列摘要

Since Taiwan is small and densely populated, motorcycles with high mobility have become a common means of transportation. Although "defensive driving" has been gradually advocated in recent years, traffic accidents have not been significantly reduced. Traffic accidents often cause casualties. The causes of traffic accidents are "failure to yield to cars in accordance with regulations" and "improper turning" accounted for the highest proportions. In addition, motorcycle drivers often compete with large vehicles for lanes. It is prone to cause traffic accidents because the blind spots of large vehicles are not easy to detect the surrounding vehicles. Recently, artificial intelligence has been applied in our daily life. In this study, an artificial intelligence system was developed using YOLO v4 deep learning technology, which has high accuracy and fast calculation speed, to detect large vehicles on the street. The image recognition features learned by machine learning were used to train and verify the test data by using the model generated by training to perform image recognition of the detection target. At the same time, it detected the distance from the surrounding vehicles. When the vehicle enters a safe distance, it will remind the locomotive driver to take preventive measures in advance through sound or light, so as to reduce the occurrence of motorcycle traffic accidents. The simulation results show that the detection accuracy of large vehicles using the AI image recognition method in this study is over 94% which is a pretty good result.