Hand Gesture is one of the most natural ways to give commands to the computer or communiate with a robot. The other plausible possibility, speech recognition, requires a lot of learning to be done on part of the computer/robot as it is user specific. In this project, we use z-depth obtained by Kinect to determine accurately different hand gestures. The advent of low cost Microsoft's Kinect has provided new opportunities & spurred research in areas of human-computer interaction (HCI).However, Hand Gesture Recognition remains a relatively unexplored facet of Kinect. The complexities involved in hand recognition owing to its small size relative to the human body & its complex shape further complicates the scenario. We use a novel distance metric method for computing dissimilarity between two different hand gestures - Finger-Earth Mover's Distance (FEMD).
We use Kinect to get the 3D depth of the image. This data is used to get the segment out the hand from the other parts of image. The contour of the hand is obtained and plotted as a histogram. This histogram is compared with other template histograms (for the predefined Gestures) and the one with the minimum FEMD values corresponds to the correct Gesture.
Area Project Proposal Presentation Final Report Code Tarball
[1] Zhou Ren, Junsong Yuan, and Zhengyou Zhang. Robust Hand Gesture Recignition Based on Finger-Earth Mover's Distance with a Commodity Depth Camera. MM' 11: 1093-1096, November 28-December 1, 2011, Scottsdale, Arizona, USA. [2] Haibin Ling, and Kazunori Okada. An Efficient Earth Mover's Distance Algorithm for Robust Histogram Comparison. ECCV' 06.