Research On The Algorithms Of Gesture Recognition Based Sparse Representation

One of the most important applications is human-computer interaction in the process of using a computer.Gesture,as a kind of natural interaction way,can make people communicate with computer directly,freely,naturally just like the way between people,therefore it has drawn the attention of researchers.There are many kinds of methods for gesture recognition.According to the method of modeling,the gesture recognition was divided into the method based on vision and the method based on wearable devices.The gesture recognition method based on visual has been research deeply.Now,those methods have got a high recognition rate through this method,however,this method relies on the external environment and crowd place too much,such as light,shade,etc.With the development of acceleration sensor in recent years,many intelligent terminals have embedded with one or more acceleration sensors,which have attract the interest of a large number of researchers.This paper mainly focuses on the gesture recognition algorithm based on 3-d acceleration sensor.Focusing on the point of users' experience in the field of human-computer interaction,gesture recognition not only need promote the gesture recognition,but also need shorten the recognition time,In this paper,the following works were done:1?In order to improve the gesture recognition rate,a gesture recognition algorithm based on sparse representation was designed in this paper,this method are divided into two phases,in the first stage,after pretreatment the acceleration data by filtering and interpolation,the KSVD method was used to train the every kind of gestures dictionary,and get the most sparse representation in the training sample over-complete dictionary,then tag every gesture sample and form a new dictionary;in the second stage,test samples were represented sparsely by the new dictionary.All kinds of samples were used to represent the test sample collaboratively,and the sparse coefficient was solved by L2 norm.The method determines the test samples to the class whose sparse coefficient is most contracted with test sample in the corresponding training samples.2?In consideration of the system requirement of real-time and high recognition rate,in this paper,a fast gesture recognition algorithm based on sparse representation was also designed.In the new method,the sparse coefficient was optimized by using L1 norm.However,the method of L1 norm optimization is time-consuming,and the runtime is proportional to the number of training sample,therefore,in this paper,the K training samples similar to the test sample was chosen firstly from the entire training samples,and then the test sample was represented by the K training samples,this method can get the sparse coefficient by the optimization method of L1 norm.When the test sample and training sample has the minimum error,the method determines the class of the training sample as the class of test sample.The improved method is better because the test sample has a closer relationship with the new training sample dictionary,therefore,it can get a higher recognition rate.Moreover the new method can reduce the runtime greatly,and improve the user experience effectively.