Gender Recognition using 3D Human Body Scans

Requirements of 3D information about the shape of the human body are growing rapidly, not only in ergonomic studies or design, but also in health-related applications. For this, a basic step is automatic gender determination. This represents an essential feature during measurement and interpretation of the data, both to perform various analyses (for example, to establish the somatotype or to assess the proportion and distribution of fat and other tissues), as well as for the appropriate diagnosis of medical conditions (evaluation of overweight, detection of malformations, etc.). Most related works deal with gender recognition by means of supervised learning based on different datasets in order to automate this task. Some focus on the analysis of facial measurements, while others focus on the processing of real-time images of silhouettes. In this work we propose a gender classifier based on analysis of 3D meshes of the human body. We describe the pre-processing techniques used, along with the supervised learning algorithms that allow solving the classification task. Finally a comparison is made between the results obtained with two different learning models: support vector machines (SVM) and decision trees.