市區監控影像之十字路口感興趣區域自動偵測與車流估計;Automatic Region of Interest Detection on Cross Roads and Traffic Flow Estimation in Urban Surveillance Videos

NCUIR > College of Electrical Engineering & Computer Science > Graduate Institute of Computer Science and Information Engineering > Electronic Thesis & Dissertation > Item 987654321/72146

Please use this identifier to cite or link to this item: http://ir.lib.ncu.edu.tw/handle/987654321/72146

Title:	市區監控影像之十字路口感興趣區域自動偵測與車流估計;Automatic Region of Interest Detection on Cross Roads and Traffic Flow Estimation in Urban Surveillance Videos
Authors:	陳俊達;Chen,Chun-Da
Contributors:	資訊工程學系
Keywords:	車流量;前景偵測;市區場景
Date:	2016-07-28
Issue Date:	2016-10-13 14:28:32 (UTC+8)
Publisher:	國立中央大學
Abstract:	車流量偵測一直是智慧型運輸系統領域中很重要的議題，研究目前實際在使用的方法，通常會在各個路口架設硬體感測器或攝影機，或是像google map取得使用者手機的GPS資訊即時統計車流量資訊。而在學術領域方面，多年來已經有許多關於影像監控的研究，針對不同場景分析各種道路資訊，例如對國道影像或是單一路口的車道場景。但是目前利用攝影機監控的作法，常常會因為攝影機架設的角度及高度，而局限住監控的範圍。因此本篇論文希望將監控的範圍拉廣，從較廣的範圍能同時獲得更多跟道路有關的影像資訊。本篇論文希望監控的範圍不僅針對單一道路，而是針對整個路口甚至多路口，希望用一台監控攝影機，即能監控範圍較廣之道路場景，並獲得各路口之車流量資訊。因此本篇論文提出一個較彈性的系統架構，自動偵測場景中十字路口的位置，並定義偵測區域，估測區域中車流量。希望攝影機架好之後便能夠自動偵測場景。因此本篇論文偵測場景中的車輛前景，並根據前景資訊尋找道路區域。藉由道路區域的輪廓獲得直線特徵，並利用直線特徵尋找可能出現路口的位置，評估每個路口的可能性。最後利用路口資訊尋找相對應的感興趣區域，作為車流量估計的範圍。並利用一個彈性的車流估計模型計算該範圍中的車流量。由於本篇論文中的監控範圍較廣，場景內的車輛較小，因此在擷取車輛資訊的過程會因為車輛過小或車輛不明顯而偵測不到該車輛，在偵測的過程我們也嘗試從感興趣區域中擷取有效的特徵資訊，以改善找不到車輛前景的情形。最後由實驗結果可以看出，本篇論文所提出的系統能夠正確的偵測場景中的十字路口區域，並在車流估計的結果也得到了不錯的誤差率，達到單一攝影機能夠偵測多路口車流之目的。 ;In this thesis, we propose a flexible traffic flow estimation system for urban surveillance videos. In recent years, there are many methods for traffic flow detection in urban scenes. However, many of them can only deal with single lane scenario. The goal of the proposed system is to automatically deal with the traffic flow estimation not only for single lane, but also for multiple crossroads. The proposed system can automatically detect the region of interests in the urban scenes. Then, the traffic flow information can be extracted from the segmented region of interest. We detect the foreground regions of vehicles and use the information to detect lane regions. Then, we find the straight lines by the contours of lane regions and use these lines to find the regions of crossroads. Finally, we evaluate these regions to choose the appropriate ones and extend these regions to define the region of interest of traffic flow estimation. Furthermore, we construct effective features in the region of interest instead of foreground extraction. The proposed method is tested on a challenging experimental dataset. Experimental results show that we can find the crossroad regions appropriately and achieve good performance of the traffic flow estimation.
Appears in Collections:	[Graduate Institute of Computer Science and Information Engineering] Electronic Thesis & Dissertation

Files in This Item:

File	Description	Size	Format
index.html		0Kb	HTML	279	View/Open

社群 sharing

Loading...