Video Component Separation Based on Compressed Robust Dynamic Mode Decomposition and Its Applications

2023 Fiscal Year Final Research Report

• Project/Area Number: 21K17767
• Research Category: Grant-in-Aid for Early-Career Scientists
• Allocation Type: Multi-year Fund
• Review Section: Basic Section 61010:Perceptual information processing-related
• Research Institution: The University of Kitakyushu
• Principal Investigator: Matsuoka Ryo 北九州市立大学, 国際環境工学部, 准教授 (40780391)
• Project Period (FY): 2021-04-01 – 2024-03-31
• Keywords: 動的モード分解 / スパースコーディング / 動画像復元 / 凸最適化

Outline of Final Research Achievements

With the advancement of digital camera technology and AI techniques, the ability to recognize objects such as people and cars in images captured by surveillance and in-vehicle cameras has rapidly evolved. Existing methods typically treat the background as static and roughly separate the moving components as the foreground. However, the camera viewpoint often moves, and noise from rain or snow streaks can be introduced, along with multiple subjects like people and cars, which have different motion characteristics.
This study proposes a video component separation algorithm that integrates dynamic mode decomposition, sparse modeling, and compressed sensing techniques to process high-dimensional data while efficiently being robust to background movement. Additionally, we explore the application of this algorithm to hyperspectral images and other ultra-high-dimensional data.

Free Research Field

Image processing

Academic Significance and Societal Importance of the Research Achievements

本研究の主要な成果は、既存の動画像成分分離技術に比べてノイズなどによる劣化にロバストな手法であるため、動画像の物体認識や動き予測の精度改善に貢献し、監視カメラ技術などの応用領域における飛躍的な技術発展が期待できる。さらに、動画像やその解析が重要な役割を果たす産業・サイエンス・工学の諸分野に大きく貢献するものである。また、ノイズなどによる劣化の観点から実現が困難であった画期的な動画像処理応用を切り開き、医療・農業・産業などの広範な分野に大きなインパクトを与えるものであると考えられる。