DEVICE AND METHOD FOR 3D RECONSTRUCTION OF HUMAN OBJECT BASED ON MONOCULAR COLOR IMAGE IN REAL TIME

• Main Authors: KWON, Yong Hoon, YOON, Ju Hong, KIM, Je Woo, PARK, Min Gyu, KANG, Ju Mi
• Format: Patent
• Language: English; French; Korean
• Published: 06.06.2024
• Subjects: CALCULATING; COMPUTER SYSTEMS BASED ON SPECIFIC COMPUTATIONAL MODELS; COMPUTING; COUNTING; ELECTRIC COMMUNICATION TECHNIQUE; ELECTRICITY; IMAGE DATA PROCESSING OR GENERATION, IN GENERAL; PHYSICS; PICTORIAL COMMUNICATION, e.g. TELEVISION

A device and method for 3D reconstruction of a human object based on a monocular color image in real time are provided. A device for 3D reconstruction of a human object according to an embodiment of the present invention extracts pixel-aligned features from a monocular image, extracts ray-invariant features from the pixel-aligned features, encodes position information of a point so as to generate encoded position information, predict an SD of the point from the extracted ray-invariant features and the encoded position information, and reconstruct a 3D human object by using the predicted SD. Accordingly, the amount of computation for SD prediction of 3D spatial points can be dramatically reduced by utilizing the ray-invariant features extracted from the pixel-aligned features and the encoded position information, such that speed can be dramatically improved. L'invention concerne un dispositif et un procédé de reconstruction 3D d'un objet humain sur la base d'une image couleur monoculaire en temps réel. Un dispositif de reconstruction 3D d'un objet humain selon un mode de réalisation de la présente invention extrait des caractéristiques alignées sur des pixels d'une image monoculaire, extrait des caractéristiques d'invariants de rayons des caractéristiques alignées sur des pixels, code des informations de position d'un point de façon à générer des informations de position codées, prédit une SD du point à partir des caractéristiques d'invariants de rayons extraites et des informations de position codées, et reconstruit un objet humain 3D à l'aide de la SD prédite. Par conséquent, la quantité de calcul pour la prédiction de SD de points spatiaux 3D peut être considérablement réduite en utilisant les caractéristiques d'invariants de rayons extraites des caractéristiques alignées sur les pixels et les informations de position codées, de telle sorte que la vitesse peut être considérablement améliorée. 단안 컬러 이미지 기반의 실시간 3D 휴먼 객체 복원 장치 및 방법이 제공된다. 본 발명의 실시예에 따른 3D 휴먼 객체 복원 장치는, 단안 이미지로부터 픽셀-얼라인드 특징을 추출하고, 픽셀-얼라인드 특징으로부터 레이-불변 특징을 추출하며, 포인트의 위치 정보를 인코딩하여 인코딩된 위치 정보를 생성하고, 추출된 레이-불변 특징과 인코딩된 위치 정보로부터 포인트의 SD를 예측하며, 예측된 SD를 이용하여 3D 휴먼 객체를 복원한다. 이에 의해, 픽셀-얼라인드 특징으로부터 추출한 레이-불변 특징과 인코딩된 위치 정보를 이용하여, 3D 공간 포인트들의 SD 예측을 위한 연산량을 획기적으로 줄여 속도를 비약적으로 향상시킬 수 있다.