Real-time hand posture recognition based on hand dominant line using kinect

Abstract

Depth sensors like Kinect can provide 2.5D depth data, which shows a extreme difference with regular color data. This paper proposes a simple and novel algorithm to perform hand posture classification using depth sensors. We present the notation of hand dominant line which can be used to make our hand descriptor rotation invariant. The recognition system combines our descriptor and SVM classifier and achieves robust hand pose recognition in real time. In cross validation test where half of dataset is used to train while the other half is used to test, we achieve 97.1% success rate on NTU Dataset and 96.2% on a subset of ASL(American Sign Language) dataset.