| DC 欄位 |
值 |
語言 |
| DC.contributor | 資訊工程學系 | zh_TW |
| DC.creator | 簡翊帆 | zh_TW |
| DC.creator | Yi-fan Jian | en_US |
| dc.date.accessioned | 2013-7-3T07:39:07Z | |
| dc.date.available | 2013-7-3T07:39:07Z | |
| dc.date.issued | 2013 | |
| dc.identifier.uri | http://ir.lib.ncu.edu.tw:88/thesis/view_etd.asp?URN=100522059 | |
| dc.contributor.department | 資訊工程學系 | zh_TW |
| DC.description | 國立中央大學 | zh_TW |
| DC.description | National Central University | en_US |
| dc.description.abstract | 智慧型運輸系統正一步步改善精進，不侷限於原本的國道影像監控，許多新硬體的研發，更是便利了交通資訊的分析，新硬體與新技術值得鼓勵，以有別於影像分析的角度著手，取得各種不同類型的資料，不僅能豐富分析的內容，擴張分析的領域，也可以輔助固有的影像資訊，達到較佳的分析效果。但是硬體開發成本遠不能及於原有的設備上深入探究其價值，所以，本篇論文將不使用新工具，僅利用已經裝設好的國道影像監控系統，在節省成本的前提下，提高影像資訊的分析精度及廣度。
利用國道影像資料分析路段車流，已有許多人付諸實行，無論是在白天或是夜間，都有許多技術的支持，即使是在天氣狀況不佳，視野不明的情況下，如：雨霧干擾。也有許多學者致力於去除雜訊，還原影像內容。此篇論文最大的特點，就在於針對不同場景下的夜間國道影像資訊，藉由顯著區域的偵測，區塊的切割，找出有利於辨別清晰或遭受雨滴汙染影像的特徵，這裡使用了統計特徵，區塊強度特徵搭配主成分分析，以及曲波係數特徵，再利用貝式分類器或支持向量機將影像作第一步的分類。若為夜間清晰影像，則採取不同的技術來估計路段車流量。反之，則進一步對受污染影像作分析，取出有用特徵，以支持向量回歸(SVR)的方式，推測出某個時間區間下經過的所有車輛數序列，進而切割序列，利用車輛數序列對應車輛數模組，推估出此路段的車流量。
本論文提出的方法架構無論在影片分類或是車流量分析都有不錯的效果，影片分類可達90%的準確度，而車流量分析誤差介於2%~3%之間，藉由估計各路段的車流量，可以即時得知路段的壅塞程度，對用路人是一大福音，每逢過年過節，國道塞車情況屢見不鮮，若將即時的分析結果與通訊裝置相配合，不僅能提供用路人由出發地至目的地的流暢行車路線，指引用路人往非壅塞路段行駛，提高各路段的平均使用率，也可作為政府實施交通管制的參考資訊。 | zh_TW |
| dc.description.abstract | In this thesis, we propose an intelligent highway surveillance application that could distinguish whether the scene is tampered by rain-drop or not and provide solutions to analyze the traffic flow under challenging rain-drop tampered conditions. There are existing methods to do traffic flow analysis in daytime and nighttime, but very few methods can deal with the conditions when the view is unclear. In the proposed system, we aim at dealing with the raining conditions at night. To deal with the difficult scenes, we extract some effective features instead of performing rain removal. We separate the system into two important parts. In the first part, we perform salient region detection and segmentation. We extract three kinds of features, including statistical features with feature selection, block intensity features with principal component analysis, and curvelet coefficients. After feature extraction, we use support vector machine and normal Bayesian classifier to do classification. In the second part, we use the same statistical features as in the first part to analyze traffic flow. We extract statistical features in detection region, and conduct support vector regression to get an estimated vehicle number for each frame. We propose a mapping model from the vehicle number sequence acquired in a number of frames to per minute traffic flow. The model depends on two factors, the length of the vehicle number sequence and the state transfer likelihood. The accuracy of traffic flow analysis is about 98%, demonstrating the proposed system is effective to analyze traffic flow at night under rain-drop tampered highway surveillance cameras. | en_US |
| DC.subject | 車流量估計 | zh_TW |
| DC.subject | 曲波 | zh_TW |
| DC.subject | 主成分分析 | zh_TW |
| DC.subject | 貝式分類器 | zh_TW |
| DC.subject | 支持向量機 | zh_TW |
| DC.subject | 坎尼邊緣檢測 | zh_TW |
| DC.subject | 顯著區域偵測 | zh_TW |
| DC.subject | traffic flow analysis | en_US |
| DC.subject | curvelet | en_US |
| DC.subject | principle component analysis | en_US |
| DC.subject | Bayesian classifier | en_US |
| DC.subject | support vector machine | en_US |
| DC.subject | SVR | en_US |
| DC.subject | canny edge detector | en_US |
| DC.subject | salient region detection | en_US |
| DC.title | 在夜間受雨滴汙染鏡頭所拍攝的影片下之車流量估計 | zh_TW |
| dc.language.iso | zh-TW | zh-TW |
| DC.title | Traffic Flow Analysis at Night under Rain-drop Tampered Surveillance Camera | en_US |
| DC.type | 博碩士論文 | zh_TW |
| DC.type | thesis | en_US |
| DC.publisher | National Central University | en_US |