Research And Realization Of Grapheme Gesture Recognition Based On Vision

In the era of rapid development of artificial intelligence,the research of computer vision is more and more popular.In the field of vision,more and more investigators begin to study on gesture recognition since the expression ability of human hand gestures is abandunt.The deep research of gesture recognition make human-computer interaction more humanized.At present the developing trend of machine is miniaturization,but the external input devices occupy a considerable part of machine.Gesture recognition which is based on computer vision make it possible to remove external input device.Currently,research on gesture recognition are mostly about simple and little gesture.In order to achieve human-computer interaction better and more simply,we use ordinary camera to constantly collect gesture pictures,and achive detection,tracking and recognition of 26 letters of an alphabet gesture,and then output the corresponding letters.The effect of gesture recognition is promoted through the analysis and improvement of the algorithm.Firstly,Skin Segmentation Detection is one of the simplest and most effective method of gesture detection.However Skin Segmentation is easy to make a detection error.For example,it is easy to mistake the human face for a human hand.On the other hand,it is inefficient to use Adaboost Classifier with Haar features to detect object on large size images.So we can take advantages of these two methods.For the detection of image,we use Skin Color Detection firstly,and put the result of detection into Adaboost Classifier,thus the precision of gesture detection is improved obviously.Secondly,Particle Filter Tracking algorithm has good results for gesture tracking.But the traditional Particle Filter Tracking algorithm is easy to particle degeneracy and particle shortage when resampling.For the disadvantage of Particle Filter Tracking algorithm,we use Wind Driven Optimization algorithm(WDO)to optimize Particle Filter Tracking algorithm.The simulation and experimental results show that the improved algorithm can improve the performance of the traditional Particle Filter Tracking algorithm.Then the results of real-time tracking are as the object identification.The recognition method is mainly based on Deep Learning algorithm—Convolutional Neural Network(CNN).For the low recognition rate and high false recognition rate of gestures which use Convolutional Neural Network,we use the Template Matching method to verity them.The overall gesture recognition rate is improved through this method.Lastly,we successfully designed a real-time gesture recognition system.This recognition system can complete gesture detection,gesture tracking and gesture recognition by collecting image via a camera,and then output the result of recognition in the form of letters.A gesture input method is realized...