識別器の特徴抽出法としての再利用による新規データセットヘの適合法

• Item type: SIG Technical Reports(1)
• 公開日: 2013-08-26
• タイトル: 識別器の特徴抽出法としての再利用による新規データセットヘの適合法
• タイトル / 言語: en
• タイトル: Reusing pre-trained classifiers as feature descriptors for adaptation to general dataset
• 言語: jpn
• 資源タイプ / 資源タイプ識別子: http://purl.org/coar/resource_type/c_18gh
• 資源タイプ: technical report
• 著者所属: 株式会社デンソー
• 著者所属: 株式会社デンソー
• 著者所属: 株式会社デンソー
• 著者所属: 株式会社デンソー
• 著者所属(英): en
• 著者所属(英): DENSO CORPORATION
• 著者所属(英): en
• 著者所属(英): DENSO CORPORATION
• 著者所属(英): en
• 著者所属(英): DENSO CORPORATION
• 著者所属(英): en
• 著者所属(英): DENSO CORPORATION
• 著者名: 坂東, 誉司; 竹中, 一仁; テヘラニ, ホセイン; 酒井, 映
• 著者名(英): Takashi, Bando; Kazuhito, Takenaka; Tehrani, Hossein; Utsushi, Sakai
• 論文抄録 / 内容記述タイプ: Other
• 論文抄録 / 内容記述: Recently, various advanced driver assistance systems have been developed, e.g., collision-warning system that detect pedestrians from frontal images captured from car mounted camera and calculate collision risk. To construct such object detector, machine-learning approach is frequently employed because of complexity of target object. However, especially in developing process, training dataset for learning the object detector cannot be fixed until final phase of developing. Although it needs quite high cost, we have to iterate gathering images and anno tating object regions to the images for learning the object detector. In this paper, we reused pre-trained pedestrian detectors, which constructed from different dataset respectively, as feature descriptors for adaptation to new dataset. Adapted detectors achieved accurate and robust detection as well as pedestrian detector constructed using the new dataset directly.
• 論文抄録(英) / 内容記述タイプ: Other
• 論文抄録(英) / 内容記述: Recently, various advanced driver assistance systems have been developed, e.g., collision-warning system that detect pedestrians from frontal images captured from car mounted camera and calculate collision risk. To construct such object detector, machine-learning approach is frequently employed because of complexity of target object. However, especially in developing process, training dataset for learning the object detector cannot be fixed until final phase of developing. Although it needs quite high cost, we have to iterate gathering images and anno tating object regions to the images for learning the object detector. In this paper, we reused pre-trained pedestrian detectors, which constructed from different dataset respectively, as feature descriptors for adaptation to new dataset. Adapted detectors achieved accurate and robust detection as well as pedestrian detector constructed using the new dataset directly.
• 書誌レコードID / 収録物識別子タイプ: NCID
• 書誌レコードID / 収録物識別子: AA11131797
• 書誌情報: 研究報告コンピュータビジョンとイメージメディア（CVIM）; 巻 2013-CVIM-188, 号 24, p. 1-6, 発行日 2013-08-26
• Notice: SIG Technical Reports are nonrefereed and hence may later appear in any journals, conferences, symposia, etc.
• 出版者 / 言語: ja
• 出版者: 情報処理学会