Paper Title
3-D Hand Pose Recognition from a Pair of Depth and Geodesic Images using Deep Convolutional Neural Network

Abstract
Accurate 3-D hand-pose recognition is one of thenovel user interface technologies that can facilitate interactions between humans and smart devices. In this work, we propose a methodology to recognize 3-D hand pose from a pair of depth and geodesic distance images of a hand. First, we train a Convolutional Neural Network (CNN) regressorwith a database of depth and Geodesic images of a hand along with the ground-truth joint positions.Second, using the trained CNN regressor, we estimate the 3-D joint positions from input pairs of depth and Geodesic images. Finally, based on the estimated joint positions, we reconstruct 3-D hand poses. Our results show that making use of Geodesic distancealongwith depthinformationimproves 3-D hand pose recognition by enhancing the capacity of regression via CNN Index Terms - Depth image, Geodesic image, 3-D hand pose recognition, Deep learning, Convolutional neural network