Identificação do somatotipo de fisiculturistas através de imagens digitais

• Ano de defesa: 2017
• Autor(a) principal: Gonçalves, Thales de Oliveira
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: por
• Instituição de defesa: Universidade Federal do Espírito Santo; BR; Mestrado em Engenharia Elétrica; Centro Tecnológico; UFES; Programa de Pós-Graduação em Engenharia Elétrica
• Palavras-chave em Português: Seleção de características; Segmentação do corpo humano; Extração automática de medições corpóreas; Fisiculturistas; Somatotipos; Engenharia Elétrica; 621.3
• Link de acesso: http://repositorio.ufes.br/handle/10/6916

Somatotype is a metric that tells us about human body shape and composition. It is important in many applications, especially in the physical education and health areas. However, obtaining the somatotype nowadays, besides being a very time-consuming procedure, demands several anthropometric devices, some of which are not very portable, and an expert of the area to take various measurements directly on the person’s body. The proposal of this work is to obtain somatotype of bodybuilders by their body images in different positions, based on image processing and machine learning techniques. Due to the difficulty of references of other works with similar proposals, a database needed to be builded by our own for the development of the proposed system. A set of measurements that are possible to be extracted from the individual’s images are proposed and a feature selection chooses a very small subset of relevant measurements to estimate the somatotype. With the assist of a segmentation technique and morphological image processings, the individual is segmented and it is proposed an algorithm to extract each of the selected relevant measurements. Finally, the body measurements taken from the individual’s images are mapped on their somatotypes based on regression techniques. The results obtained shows that the somatotype of bodybuilders can be estimated reasonably based only on their images, which is a less expensive option to obtain this metric.