3D reconstruction of human bodies from single-view and multi-view images: A systematic review

• Published in: Computer methods and programs in biomedicine 2023-09, Vol.239, p.107620-107620, Article 107620
• Main Authors: Correia, Helena A., Brito, José Henrique
• Format: Article
• Language: English
• Subjects: 3D Representations; Human bodies 3D reconstruction; Human Body; Humans; Imaging, Three-Dimensional - methods; Multi-view 3D reconstruction; Single-view 3D reconstruction; Systematic review; Virtual Reality
• ISSN: 0169-2607; 1872-7565
• DOI: 10.1016/j.cmpb.2023.107620

•The recent methods for 3d reconstruction of human bodies are categorized and analyzed.•A comprehensive description of each methods is provided.•Identifying of topics for further research.•A successful 3D reconstruction approach generally depends on the information extracted from the image (structural, shape, and texture information) and the choice of the output 3D representation. [Display omitted] 3D reconstruction of human bodies remains an important research issue in computer vision. It is a fundamental resource for many everyday applications, including in the entertainment industry (animation), human-computer interaction, medical diagnosis, and, more recently, virtual reality. In recent years, several researchers have proposed multiple approaches for 3D reconstruction of human bodies from images. This article aims to display a systematic review of the different methods developed for 3D reconstruction of human bodies from single or multiple images. With this work, we intend to explore and categorize the different methods of 3D reconstruction of human bodies, making a comparison and then a discussion between the approaches. For that, articles published between 2016 and 2023 were analyzed. A total of 3325 publications were identified in the specified databases, and after a detailed analysis, only 70 were reviewed for this systematic literature review. Based on the analysis performed, it is concluded that it is possible to categorize human 3d reconstruction methods into two principal groups that are divided according to their theoretical category or based on the 3D representation used as output. However, more research should be done on human inference to overcome existing limitations and generate more accurate and detailed models.