以機器學習為基礎之自動化車流影像辨識技術

蔡曉佩 ； 黃仲誼 ； 曾建澔 ； 游寶達 ； 林芳邦

《TANET2017 臺灣網際網路研討會》 2017年 (2017/10) Pp. 868-870

https://doi.org/10.6728/TANET.201710.0150

摘要

由於臺灣地區經濟快速成長，運輸路網發達、車輛普及化，進而衍生許多交通相關問題。根據統計資料顯示，台灣地區人口已達2355.2萬人，另汽機車登記數量共計2160.4萬輛，平均每1人就擁有一部汽機車，造成汽機車數量過大而導致許多民生問題，包括交通擁塞、都會區車輛排氣污染等，更成政府施政的重點之一。因此車流量的估測與計算是一個很重要之研究因素，我們可以透過車流量區分路段的尖峰與離峰時間，進而以替代道路或是調撥車道等其他方式紓解交通擁塞的問題；亦可透過車流種類分類運用至運輸交通影響空汙研究調查，進而探討影響該區域空氣汙染的原因。由於目前物聯網技術的快速發展，加上路口及交通監視器的廣泛普及，利用攝影機監測交通狀況，並搭配自動影像處理分析進行即時交通資訊萃取，將可大幅度的省下許多硬體設備的花費，提升交通相關問題的處理綜效，所得的資料並可進行大資料分析，極具其他後續研究價值。

關鍵字

車流量計數 ； 車輛影像辨識 ； 卷積神經網路

並列摘要

The high growth of economy and transportation network in Taiwan introduces a very high density of vehicles. Taiwan's population is now 23.55 million, while the number of automobiles and motorcycles is more than 21.6 million. It means an average of 0.917 vehicles per person. These enormous amounts of vehicles have caused traffic congestion, vehicle exhaust pollution and many other problems in the urban areas, which quickly became the top priority issues to the government nowadays. The estimation and calculation of traffic flow are the important research factors. We can distinguish the peak and off-peak time of the traffic through the traffic flow, and then alleviate the traffic congestion in alternative routes or other ways. The reasons of the air pollution in the specific regions could also be explored by the classification of the different traffic flow and investigations on pollution effects of transportation. With the rapid development of the IoT technology and the wide spread of the traffic camera, the capability to automatically extract vehicle information from the video is extremely invaluable. These data can serve as the foundation to help solving the traffic related problems in a very cost-effective way. Furthermore, their use in the follow up big data research will be significant.

並列關鍵字

Traffic flow counting ； Vehicle Image Recognition ； Convolutional Neural Network