3D-Aware Generalized Instance Segmentation for AI-based Video Surveillance of Smart City Roads

カテゴリ: 研究会(論文単位)

論文No: CMN24026

グループ名: 【C】電子・情報・システム部門 通信研究会

発行日: 2024/03/25

タイトル(英語): 3D-Aware Generalized Instance Segmentation for AI-based Video Surveillance of Smart City Roads

著者名: Joshi Sharad(Hitachi India Pvt. Limited),Ganesh Ananth(Hitachi India Pvt. Limited)

著者名(英語): Sharad Joshi(Hitachi India Pvt. Limited),Ananth Ganesh(Hitachi India Pvt. Limited)

要約(日本語): Hitachi Ltd has been building state-of-the-art computer vision solutions for manufacturing, industrial solutions, smart cities etc. AI-based visual analysis of real-world objects’ images/videos acquired using cameras is essential for various applications related to surveillance, security, and fault detection. Instance segmentation is a basic computer vision task which facilitates object-level analysis. We present a solution for generalized instance segmentation which can generalize segmentation for a larger set of categories unseen during training, unlike traditional instance segmentation methods. The proposed 3D-aware solution scales for unseen categories while saving on annotation and training costs.

要約(英語): Hitachi Ltd has been building state-of-the-art computer vision solutions for manufacturing, industrial solutions, smart cities etc. AI-based visual analysis of real-world objects’ images/videos acquired using cameras is essential for various applications related to surveillance, security, and fault detection. Instance segmentation is a basic computer vision task which facilitates object-level analysis. We present a solution for generalized instance segmentation which can generalize segmentation for a larger set of categories unseen during training, unlike traditional instance segmentation methods. The proposed 3D-aware solution scales for unseen categories while saving on annotation and training costs.

本誌: 2024年3月28日-2024年3月29日通信研究会

本誌掲載ページ: 49-53 p

原稿種別: 英語

PDFファイルサイズ: 1,506 Kバイト