Name: GENTIL AUER NETO

Publication date: 15/06/2020
Advisor:

Namesort descending Role
PATRICK MARQUES CIARELLI Advisor *

Examining board:

Namesort descending Role
JORGE LEONID ACHING SAMATELO External Examiner *
KLAUS FABIAN COCO External Examiner *
PATRICK MARQUES CIARELLI Advisor *

Summary: The human body can be referenced by a metric that measures up shapes and proportions: the somatotype. Studies shows that its identification is useful in several areas of health and physical education. Currently, the main method for its estimation requires a lot of time, specialized training and the use of anthropometric equipment for measuring body parts. Some researches are looking for simplifying the exam, for example, by automating it with digital image processing. However, there are limitations, such as the adaptation
of environments for capturing good quality photographs, an important requirement for its viability. Considering the benefits of obtaining a robust and fast method, such that simplifies the exam realization and reduces its costs, the main purpose of this work is to research and develop methods that are able to automatically predict information from the human body through images obtained in environments without previous preparation. In this context, this work proposes to combine machine learning techniques, in particular deep learning, in order to achieve detailed segmentation of the body contour of individuals in uncontrolled environments. In addition, gender identification and weight estimation are also done, both by facial analysis. Four different datasets are used: face2bmi, Wiki, Fisio-Somatotipo and Internet. The two first containing face images of people in general and the last two with photos of full bodies of athletes. The results shows that the study is promising, having achieved a good level of segmentation in adverse environments and 91.52% of accuracy for gender classification. It is also discussed how gender can influency
on the weight prediction model, which reached a mean error of 5.56 kg as best result.
Keywords: Somatotype. Deep Neural Network. Human body segmentation. Uncontrolled Environments. Gender Classification. Body Weight Estimation.

Access to document

Acesso à informação
Transparência Pública

© 2013 Universidade Federal do Espírito Santo. Todos os direitos reservados.
Av. Fernando Ferrari, 514 - Goiabeiras, Vitória - ES | CEP 29075-910