Study On Vision-based Hand Gesture Recognition

Nowadays, scholars, domestic and overseas, are devoting themselves in hand gesture recognition field, which has brought so many conveniences to people’s life. However, on account of the complication of hand gesture recognition system, so many problems works still need to be done. Thus this paper focuses on the filed of hand gesture recognition algorithms. The paper has built two hand gesture recognition systems. One of them is hand gesture recognition system based on SVM and the other one is hand gesture recognition system based on Deep Learning.On the basis of hand gesture recognition system, the contribution of this thesis is as follows:Firstly, this thesis use median filter algorithm, Gray World color balance algorithm, Gaussian Mixture Model in YCgCr space algorithm, adaptive threshold binarization algorithm and Laplace operator algorithm to deal with hand gesture images, which has achieved accurate segmentation result. This has laid a very firm foundation for our future study.Secondly, this thesis proposes the method of hand gesture detection by combining frame difference algorithm with skin color to satisfy the requirement of accurate and real time. After detection the paper use Camshift algorithm to tracking hand to reduce the interference from surrounding. The detection and tracking algorithm effectively reduce the calculation and improve the performance of the whole system.Thirdly, the thesis establishes a hand gesture recognition system based on SVM. In this system, Hu invariant moment, Fourier descriptor and hand gesture area feature are used as the eigenvector of hand gesture. The SVM system has achieved very high recognition rate. At the same time, in order to compare with Deep Learning system, the paper also has test the recognition rate while only Hu invariant moment is used as eigenvector.Lastly, the thesis has applied Deep Learning algorithm into hand gesture recognition system. This system also has two different models. One of them directly use binarized hand gesture images as input of Deep Learning system. The other one use Hu invariant moment as input. Then compare the experiment result with SVM system, respectively.