夜間即時影像處理擷取交通參數之研究

Abstract

隨著ITS之發展，交通資料之搜集亦日益重要，在過去偵測設施上，多利用單價便宜、好用之環路線圈進行車流參數收集，乃至於微波、雷達、紅外線等偵測器之發展，而近年來則是朝向影像式偵測器之開發，影像式偵測器本身可搜集之交通參數多過於其他種類偵測器，且建置簡單又不破壞原有道路鋪面，因此其未來發展性極大，目前在市面上之影像式偵測器皆為套裝模組式，因此其價格較昂貴，且可能不符合地方交通特性，是以近年來許多研究便朝向自行開發演算程式，以對車流影像進行交通參數擷取。 又在影像式偵測器相關研究上，主要多以白天影像進行探討，鮮少專門針對夜間影像做分析探討，故本研究將針對夜間影像進行處理；又過去研究多利用固定式攝影機擷取影像，但隨著網路之快速發展，現下越來越多交通資訊網頁可直接察看路側CCTV之影像，因此本論文不同於以往研究，乃是以網路傳輸式之CCTV影像作為分析對象，又網際網路傳輸式影像，其品質深受網路傳輸速度及頻寬之影響，當網路傳輸速度減慢之時，可能造成影像之停頓與跳格等現象，且網路式影像偵測技術，其攝影機角度為交控人員為監看路側情況所調整設定，因此無一定之攝影角度，故演算法須考量更多之情境。 本論文以網路式攝影機夜間影像為對象進行研究，其傳輸速度約為10fps，利用Java程式語言進行影像處理，並以佈設車道線、車流計數器為主要方法進行交通參數擷取，為了適應網路式影像之特性，可以車流計數器計算車流量，並依壓占率反算車流平均速度，亦可以抽樣擷取式追蹤車輛區塊演算車流平均速度，本論文最終目的為增加CCTV功能性，開發即時影像處理技術。

With the development of ITS, the Traffic Monitoring System is more and more important. Between the many kinds of the vehicle detectors, the computer vision-based vehicle detector has more excellence than other, because that it can obtain more traffic parameters. The processes of daytime image and night image are very different. So we wish to build up a vision-based vehicle detector aimed at the night vehicle flow image. In this study, we focus on how to extract traffic parameter from instant images at night via image processing. We use the roadside CCTV to capture the image and transmit it through Ethernet. And the image processing procedure includes image pre-processing, background building, morphology processing, camera calibration, sampling tracking, traffic parameter extraction. Before this system start to work, the administrator can set the land line, the image collecting area, and the vehicle flow counter depending on the image circumstances. Hence we can deal with more situations and make the technique more elastic. Because the night image has more problems than daytime image and deeply influenced by the illumination so we should overcome and adjust more difficulties. Then we establish a platform and use friendly user interface; thus people can operate it easily. And this real-time networked image detector can be used to process the online CCTV image. Besides, the experimental result shows that we can obtain the vehicle volumes, velocity, occupancy, and etc. And the accurate rate of the vehicle volume can achieve 90%. And the velocity also has enough reliability. The study gives some conclusions and discussions in the end.