Analysis of Object Shape and Illumination from Image Sequence Using Unconventional Models

2023 Fiscal Year Final Research Report

• Project/Area Number: 20K11866
• Research Category: Grant-in-Aid for Scientific Research (C)
• Allocation Type: Multi-year Fund
• Section: 一般
• Review Section: Basic Section 61010:Perceptual information processing-related
• Research Institution: Okayama University
• Principal Investigator: Migita Tsuyoshi 岡山大学, 環境生命自然科学学域, 助教 (90362954)
• Project Period (FY): 2020-04-01 – 2024-03-31
• Keywords: パラメータ推定 / 画像列 / 非線形最適化 / ヤコビ行列 / レイトレーシング

Outline of Final Research Achievements

In this research, we have developed a method for estimating parameters, such as object shape and illumination, from a given set of images. The formulation is
minimization of the differences between given and synthesized images. The synthesys employs the ray-tracing algorithm, which enables us to exploit reflections and shadows, which are often considered to be harmful for estimation, in estimating shape and/or illumination parameters. The method also make use of multiple viewpoints. To do so, a new image representation model is introduced, which enables us to efficiently calculate the image and its gradient with respect to locally-affected parameters, i.e. texture intensity values. This is combined with difference-quotient-based approximation w.r.t. globally-affected parameters, i.e. object shape and light positions, to construct a joint estimation method.

Free Research Field

コンピュータビジョン

Academic Significance and Societal Importance of the Research Achievements

画像に映った物体は，カメラや光源の位置によって見え方が大きく変化するため，これらの影響を取り除くことが，物体の形状や色彩を正確に推定する上で重要である．これが可能であれば，画像からの3次元モデル生成（写真測量）や人体の状態の解析等に利用でき，HCI，個人認証，自動運転等の基礎としても有用である．
本研究では，近年普及が進むハードウエア支援レイトレーシング法と深層学習系の手法（ともにGPUを利用する）を組み込んだ画像解析を検討し，更なる研究の基礎となる技術を確立した．