Identificação biométrica com antropometria e caminhar humano utilizando o Kinect

Abstract

The present work shows the result of applying machine learning algorithms to classify individuals by anthropometric attributes and human gait obtained with the help of Microsoft Kinect sensor as an alternative to obtaining a model of the human body during walking. To investigate the use of Kinect in this task, the methodology applied in the use of 3D positions of the major joints tracked by the sensor, or the “skeleton” of 140 individuals captured for this work, walking in front of the Kinect is presented. From these 3D points, human gait and anthropometric parameters were obtained. It explored the classification taking into account the number of individuals, the number of frames required for classification by anthropometry, and the number of gait cycles captured towards a satisfactory accuracy using attributes of human gait. The results obtained are compared with the number of correctly classified instances provided in works state-of-art, and show that the proposed methodology provides superior or similar results to those provided in the referenced work. Finally, additional experiments are also presented, such as sorting by gender (male and female) and body mass index.