Detección automática de manos con caminos geodésicos en datos multi-modales

Abstract

In the last years, with the appearance of the multi-modal RGB-Depth information provided by the low cost KinectTM sensor, new ways of solving Computer Vision challenges have come and new strategies have been proposed. In this work the main problem is automatic hand detection in multi-modal RGB-Depth visual data. This task involves several difficulties due to the changes in illumination, viewport variations and articulated nature of the human body as well as its high flexibility. In order to solve it the present work proposes an accurate and efficient method based on hypothesis that the hand landmarks remain at a nearly constant geodesic distance from automatically located anatomical reference point. In a given frame, the human body is segmented first in the depth image. Then, a graph representation of the body is built in which the geodesic paths are computed from the reference point. The dense optical flow vectors on the corresponding RGB image are used to reduce ambiguities of the geodesic paths’ connectivity, allowing to eliminate false edges interconnecting different body parts. Finally, as the result, exact coordinates of both hands are obtained without involving costly learning procedures.