Research On Gesture Recognition Based On Kinect

With the rapid development of cloud computing,block chains,Internet of things,and artificial intelligence,the way people interact with machines is constantly redefined.People are actively exploring more intelligent ways of interaction,and gesture recognition is an important part of interaction.Gesture recognition means that people can use simple gestures to control objects or interact with machines,so that computers can understand human behavior.At present,gesture recognition is mainly based on visual processing method,which is greatly affected by external factors such as illumination.The effect of separation and detection of hands is not ideal.A Kinect sensor of Microsoft company was used to study gesture recognition in the paper.The hand joints were located by skeleton tracking method,so hand regions was separated.Two gesture recognition algorithms of Support Vector Machine(SVM)and Back Propagation(BP)neural network were designed.The BP algorithm was optimized by Genetic algorithm to improve the recognition rate to 97.59%,which was 2.2% higher than that of BP algorithm.The main tasks were described as following:(1)The related theories,advantages and disadvantages of gesture recognition methods at home and abroad were analysed.The depth imaging principle of Kinect sensor and its advantages in human skeleton tracking were analysed.(2)The region separation,image processing and extraction of the hand contour were based on the acquisition images of Kinect.Based on the spatial depth information,the region of the front and back of the hand were separated and only the image of the depth of the hand was retained by the region separation method,so that the recognition was not almost affected by the environment such as background illumination.Images were processed by the method of median filtering,corrosion and expansion.The hand contour was extracted by Freeman chain code.(3)Image invariant moments were used as the feature of gesture recognition,which has the invariance of translation,scaling and rotation.Support Vector Machine algorithm and Back Propagation algorithm were designed to recognize the hand gesture.Because the Gauss kernel function has many advantages,such as the fast learning speed,and good learning characteristics in the case of small number and low dimensions of samples,it was chosen in SVM classifier.The parameters of SVM and BP algorithm for gesture recognition were determined through experiments.In order to improve the recognition effect,a neural network optimized by genetic algorithm was proposed to obtain the optimal weights and thresholds,which solved the problems of slow convergence time and existence of local minimum.In summary,a large number of sample images were sampled by Kinect.Median filter and morphological processing method were applied to image processing.Two gesture recognition algorithms of SVM and BP were designed and the BP algorithm was optimized based on genetic algorithm.The effect of gesture recognition was evaluated by the accuracy,real-time and stability of gesture recognition.The results showed that the gesture recognition studied in this work was of practical significance,and the gesture recognition rate and efficiency can be raised effectively by the gesture recognition algorithm of this paper.