Co-occurrence Probability-based Pixel Pairs Background Model for Robust Object Detection in Dynamic Scenes

カテゴリ: 部門大会

論文No: TC6-3

グループ名: 【C】平成25年電気学会電子・情報・システム部門大会講演論文集

発行日: 2013/09/04

タイトル(英語): Co-occurrence Probability-based Pixel Pairs Background Model for Robust Object Detection in Dynamic Scenes

著者名: Dong Liang(北海道大学),金子 俊一(北海道大学),橋本 学(中京大学),岩田 健司(産業技術総合研究所),Xinyue Zhao(浙江大学),佐藤雄隆 (産業技術総合研究所)

著者名(英語): Dong Liang(Hokkaido University),Shunichi Kaneko(Hokkaido University),Manabu Hashimoto(Chukyo University),Kenji Iwata(National Institute of Advanced Industrial Science and Technology (AIST)),Xinyue Zhao(Zhejiang University),Yutaka Satoh(National Institute of Advanced Industrial Science and Technology (AIST))

キーワード: object detection|sudden illumination fluctuation|burst motion|background modeling|co-occurrence probability

要約(日本語): An illumination-invariant background model for detecting objects in dynamic scenes is proposed. It is robust in the cases of sudden illumination change and burst motion. Unlike the previous works, it distinguishes objects from a non-stationary background using co-occurrence character between a target pixel and its supporting pixels in the form of multiple pixel pairs. We use a two-stage training framework for modeling the background. Joint histograms of co-occurrence probability are employed to screen supporting pixels with high correlation coefficient values; the background model maintains a sensitive criterion with few parameters for detecting foreground elements. Experiments using several challenging datasets proved the robust performance of object detection in various environments.

PDFファイルサイズ: 7,366 Kバイト