Real-time Vision-based Hand Gesture Recognition Using Sift Features

Abstract

This paper introduces a new algorithm based on machine vision for the recognition of hand gesture. In step 1, the Microsoft Kinect sensor is used to capture color images and depth. User’s hand detected by eliminating the background and rescaling image. In the next step, “Scale-invariant feature transform (SIFT)” algorithm is used for the feature extraction. The extracted feature vectors are built in vocabulary tree with K-means clustering. Finally, Hand gesture is recognized by a recognition high-level method called stochastic context-free grammar (SCFG). SCFG is used for syntactic structure analysis that is based on hand gesture recognition, that is combined postures can be analyzed and recognized by a set of production rules. SCFG is most effective in disambiguate. By this approach, we are able to recognize various gestures in 30 frames per seconds (fps) and with more than 90 % accuracy.

DOI: http://dx.doi.org/10.11591/telkomnika.v15i1.8091