Hand Segmentation by Fusing 2D and 3D Data

Abstract

This paper describes a robust technique based on the fusion of 2D data and 3D information for hand segmentation. 2D data is the closed region delineated by the boundary and the color information. 3D data is provided by estimating the disparity map using images from two apart cameras. The disparity map is used for generating an intensity image with a rough range estimation for each pixel. The color and range information are fused as the input data for the segmentation algorithm. Our proposed segmentation technique is based on the Gaussian Mixture Model (GMM) which helps us to determine the hand region pixel clusters in the fused data. The experimental results show that the proposed segmentation technique can successfully segment the hand from users body, face, arm or other objects in the scene under variant illumination conditions in real time.