A real time fingers detection by symmetry transform using a two cameras system

Abstract

We propose a two cameras system for real time fingers detection: one camera aims at user's face and is used to build its skin color model, whereas the second camera is focused on his hand. The face detector is a cascade of boosted classifiers, and it is run periodically across time to update the operator skin color model. Color histogram of face candidate is build, and backprojected into the frame of the second camera resulting in a skin color probability map. A region of interest is defined by Camshift tracking of the hand: gradient field of the skin probability map is computed and orientation of edge points is determined. Oriented edges are subject to a symmetry transform, a cumulative approach called Chinese Transform (CT), which outputs an accumulator highlighting the axis of symmetry of the image. Post-processing of the CT accumulator results in fingers detection and localization: application to pointing and grasping gesture is illustrated. The two cameras are processed simultaneously in an average of 50 frames/s on a standard PIV PC. © 2008 Springer Berlin Heidelberg.