UTokyo Repository 東京大学
 

UTokyo Repository >
124 情報理工学系研究科 >
40 電子情報学専攻 >
1244025 修士論文(電子情報学専攻) >

このページ(論文)をリンクする場合は次のURLを使用してください: http://hdl.handle.net/2261/52456

タイトル: Adherent raindrops detection and removal in video
その他のタイトル: 動画像におけるレンズ上の雨滴の認識と除去
著者: You, Shaodi
著者(別言語): 尤, 少迪
発行日: 2012年9月27日
抄録: Raindrops appeared on windscreens or window glass can degrade the visibility of the outside scenes. If we can detect and later remove the raindrops, many applications such as intelligent vehicle system will benefit from it. In this thesis, we intend to focus on developing a method of automatic raindrop detection. We mathematically analyze the characteristics of raindrops adhered to a clear transparent surface, particularly the characteristics of the motion and the changes of the intensity values. The analysis is motivated by the fact that a raindrop conceives the information of the whole background scene, behaving like a minuscule catadioptric lens. In conducting the analysis, we found that motion of raindrop pixels is slower than that of non-raindrop pixels, and the change of intensity of raindrop pixels is smaller than that of non-raindrop pixels. Based on these findings, we developed our detection method. Experimental results using real videos show the effectiveness of our method. Having detected raindrops, in this thesis, algorithms to automatically repair the video in complex outdoor environment are proposed. We categorize outdoor environments into 3 situations and propose a video repairing algorithm for each of them. First, for a fast motion, structured and textured video, a sparse matching based algorithm to estimate dense motion is described. A classification based on the dense motion is introduced and raindrop removal are proposed. Second, for a slow motion, structured and textured video, a raindrop removal method using image registration is introduced. Third, for a static and non-structured video, a modified image inpainting method is introduced for raindrop removal. Experimental results using real videos show the effectiveness of our method.
内容記述: 報告番号: ; 学位授与年月日: 2012-09-27 ; 学位の種別: 修士 ; 学位の種類: 修士(情報理工学) ; 学位記番号: ; 研究科・専攻: 情報理工学系研究科電子情報学専攻
URI: http://hdl.handle.net/2261/52456
出現カテゴリ:025 修士論文
1244025 修士論文(電子情報学専攻)

この論文のファイル:

ファイル 記述 サイズフォーマット
48106453.pdf10.73 MBAdobe PDF見る/開く

本リポジトリに保管されているアイテムはすべて著作権により保護されています。

 

Valid XHTML 1.0! DSpace Software Copyright © 2002-2010  Duraspace - ご意見をお寄せください