3D Human Body Shape Analysis Using Deep Learning and Computer Vision Techniques

Obesity and overweight are complex chronic diseases whose causes and potential interventions are widely understood. However, in recent decades, the number of cases has grown alarmingly, affecting both young people and adults worldwide. Despite advances in treatment strategies, many nutritional programs fail due to a lack of adherence by patients. This lack of commitment hinders patients’ active involvement, limits their awareness of the severity of their condition, and ultimately compromises treatment outcomes. This thesis presents a set of methods and tools designed to improve the quality of nutritional treatments and help patients increase adherence to their treatments, thus achieving their goals. The primary focus of this thesis lies in the creation of a methodological framework that enables the acquisition, processing, and analysis of three-dimensional (3D) models of the human body obtained through RGB-D cameras, as well as clinical data collected by nutritional specialists. For data acquisition, a camera system was designed and validated to capture precise 3D models of the human body at different stages of treatment. These models allow for both visual and quantitative evaluation of body morphology changes over time, providing detailed information that facilitates clinical analysis and motivates patients. Additionally, this thesis addresses the processing of clinical and 3D data using computer vision and artificial intelligence techniques. The developed tools enable the standardization and enhancement of 3D models of patients, leveraging parametric human body models. This advancement has led to the creation of a multimodal database, allowing for the correlation of 3D models with clinical data. Using this database, analyses have been performed with deep learning techniques. A system was developed to synthesize all data into a latent space, enabling patient progress visualization, clustering, and predictive modeling. Furthermore, a neural network was proposed to generate a 3D human body model solely from clinical data. The results demonstrate the potential of applying computer vision techniques to address clinical challenges, improving both the processes of nutritional treatments and the patient experience. The contribution of this work opens new opportunities at the intersection of technology and health, laying the groundwork for future research in the personalization of medical and nutritional treatments.

Author

• García-D'Urso, Nahuel

• 3D Model
• Human Body
• Variational Autoencoder
• Machine Learning
• Anthropometric Measurements
• Nutritional Treatment

Identifier

hdl:10045/163082

OAI identifier oai:identifier

oai:rua.ua.es:10045/163082