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/2261/52235

タイトル: Head Pose Estimation from Low Resolution Image with Scene Adaptation
その他のタイトル: 環境への自動適応を伴う低解像度画像からの頭部姿勢推定
著者: Chamveha, Isarun
著者(別言語): ジャムウェイハー, イサラン
発行日: 2012年3月22日
抄録: Head pose estimation technique is a core of many computer vision applications. Head pose estimation is often used as cues to estimate human attention. Due to limitations such as camera placement location or system resources, it is not always appropriate to install high resolution cameras on every system. Head pose estimation from low resolution images are desired in such situation. Although many methods have been proposed for head pose estimation from low resolution images, there are many technical limitations for the implementation of the system. To construct an effective head pose estimation system, a large number of head pose training samples collected from the same environment as test environment are desired because the process of acquiring such dataset is extremely time-consuming and makes it impossible to prepare scene-specific dataset for every scene. The first part of this paper describes the method to automatically obtain scene-specific dataset. This method exploits the observation that people are more likely to to turn their head to where they are walking. With this observation, the tracking method is applied to the video taken from the scene beforehand. Head pose training data are then inferred from tracking results and are acquired automatically. This method enables automatic acquisition of training data and solves the problem of data collection. The second part describes the method to improve head pose estimation in scenes with short available videos or low number of walking pedestrians so adequate amount of head pose samples cannot be captured. Datasets captured from various scenes are used to alleviate the problem and increase the accuracy significantly. The third part describes the method which improves head pose estimation accuracy for scenes with dramatic difference within the scene. This method adaptively divides scenes into multiple parts in order to localize head pose estimator.
内容記述: 報告番号: ; 学位授与年月日: 2012-03-22 ; 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/2261/52235
出現カテゴリ:025 修士論文
1244025 修士論文(電子情報学専攻)


ファイル 記述 サイズフォーマット
48106416.pdf11.38 MBAdobe PDF見る/開く



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