Gesture Recognition And Implementation Based On Convolutional Neural Network

With the development of computer technology and the Internet,human-computer interaction technology has made great progress and became rich in content.Finding a simple and efficient way for human-computer interaction has become an urgent problem.Language,body movements and expressions can express human intentions in daily life.But Gestures are currently a hot research because of their easy understanding.There are three key steps in gesture recognition,including image preprocessing,gesture segmentation,and gesture feature extraction.In order to obtain better gestures images and avoid the influence of noise from real world such as lighting,the white balance technology which can corrected errors caused by lighting and equipment factors is discussed in detail.The Mean Blur,Gaussian Blur and Median Blur on image noise elimination are also introduced in this paper.Another critical step is to realize the processing of image morphology,which lays the foundation for the following gesture segmentation steps.The method based on combining a depth image with a color image,and a method based on skin information are frequently used methods for gesture segmentation.As the feature of color,skin color has better stability than other appearance features.Skin color can show strong anti-interference ability in a complex background.This article not only reviews three color spaces for skin color segmentation technique,but also establishes skin color models in the RGB,HSV,and YCbCr color spaces,respectively and compares them experimentally.YCbCr color space can be quickly converted from the RGB color space and behaves well so finally is selected for skin color-based gesture segmentation.On the other hand,the feature extraction of the gesture features makes full use of the self-learning of the convolutional neural network,and the neural network autonomously learns and extracts the features of the gesture to complete the image recognition.After comparing multiple public databases,a database of 26 gestures is established and used for neural network training.The accuracy of the network parameters can achieve 99.9%.This paper uses Qt as a graphical user interface development tool for the system.It is mainly composed of gesture acquisition,gesture processing,and gesture recognition.By clicking on the corresponding button to obtain gestures from different sources to complete the recognition,the system has higher accuracy and real-time performance...