View-Invariant 3D Hand Trajectory-Based Recognition

This paper proposes a novel method for view-invariant 3D hand trajectory-based recognition. The image depth information in gesture segmentation is collected by using Kinect sensor. View-invariant 3D hand trajectory is represented by improving centroid distance function. Hidden Markov model is applied to train and recognize hand gesture. Experiment results show that the proposed method is robust under the condition of different illumination and complex background. The illustrated system can successfully recognize spotted hand gestures with a 97.7% recognition rate for Arabic numbers 0 to 9.