Method For Segmentation Of Articulated Structures Using Depth Images for Public Displays
Robin Watson
Department of Computer Science and Software Engineering
University of Canterbury
Abstract
A novel method is presented to analyse articulated structures in depth data and is used in an attempt to implement gesture-motion control. The method first uses region growing with a depth threshold to obtain an initial segmentation of the scene into different bodies. Region growing is carried out again on these bodies to produce subregions. A head tracking method and hand tracking method were implemented using the depth analysis. The head tracking had an average of 22 pixel error. The hand tracking was unsuccessful.