Supervised, +, Unsupervised Classification for Human Pose Estimation with RGB-D Images: A First Step Towards a Rehabilitation System

Abstract

A system has been developed to detect postures and movements of people, using the skeleton information provided by the OpenNI library. A supervised learning approach has been used for generating static posture classifier models. In the case of movements, the focus has been done in clustering techniques. These models are included as part of the system software once generated, which reacts to postures and gestures made by any user. The automatic detection of postures is interesting for many applications, such as medical applications or intelligent interaction based on computer vision.