Hand gesture recognition based on skeletal image properties

Abstract

Hand gesture Recognition is an important task and can be used in a lot of applications. In intelligent systems, hand gesture recognition can be used to access information through a video interface. In recent years, skeleton-based hand gesture recognition become a popular research topic. The existing methods have the low discriminative power due to sensitivity of features to image noise. We have proposed new methods to decrease the influence of the noise to extract hand image features. The objective of the research is to improve the hand gesture classification accuracy. A hand gesture recognition method based on skeleton image properties is developed. For 5 classes recognition, this approach allows us to increase the classification accuracy on test set from 0.4% till 20.4% as compared with existing well-known methods.For 10 classes recognition, this approach allows us to increase the classification accuracy from 5% till 18% as compared with existing well-known methods.