Research On Error Detection And Correction Algorithms Of Misunderstanding Gestures Based On Convolutional Neural Network

With the vigorous development of machine vision technology,gesture recognition technology plays an increasingly important role in human-computer interaction with its convenient way of body language communication,and is loved by the majority of human-computer interaction researchers.Thanks to the growing maturity of gesture recognition technology,intelligent education,intelligent medical care,intelligent social security and other intelligent systems with gestures as interactive tools are gradually popularized in people's daily life.Therefore,this paper constructs the gesture database based on the semantic flexible mapping interaction model.Therefore,this paper builds a gesture database based on the semantic flexible mapping interaction model.However,in the practical application of the gesture model using convolutional neural network,there are some similar gestures,which will inevitably reduce the performance of the whole system.But the intelligent system with poor universality and poor effect will be difficult to be accepted and adopted by people.Based on the support of the National Key R&D Program of China and the National Natural Science Foundation of China,the Shandong Provincial Key R&D Program,this paper studies confused gestures in teaching process using intelligent interactive interface as application platform.The main research goal of this paper is to e xplore the error mechanism of mistaken gestures based on convolution neural network gesture recognition algorithm,and to propose an intelligent error correction algorithm based on probabilistic statistical model and convolution feature,and to realize the intelligentization of interactive teaching through intelligent detection and error correction of mistaken gestures.The main research contents and innovations of this paper are reflected in the following aspects:(1)Even if the gesture recognition algorithm is very mature,but in the actual application due to external factors interference,it is inevitable that there will be false gestures.Compared with the traditional static gesture recognition algorithm based on convolutional neural network,this paper proposes a probability model of misjudgement based on probability statistics,which can realize intelligent error correction of misjudged gestures.This algorithm mainly analyses the large data,establishes the confusion matrix of misrecognition gestures,and analyses the quantitative relationship between each type of gestures and other types of gestures misidentified by the gesture recognition model.Based on the predicted recognition results and the probability function of the actual categories,a misjud gment probability model is established.After detecting the wrong gestures,this paper uses the probability matrix of misjudgment as the base point to generate random numbers,and designs a probability generator to correct the errors,which improves the recognition rate of the confused gestures by about 5%.(2)The existing methods generally improve the recognition rate by optimizing the network structure and training parameters,or by defining new gesture features and similarity operators to improve the recognition rate.In contrast,this paper explores the error mechanism of gesture misrecognition and seeks for the regularity of error process.(3)Since the above probability-based error correction algorithm does not reflect the gesture error process,and the error correction effect is not ideal,this paper proposes an intelligent gesture error detection and error correction algorithm based on convolution feature.In order to explore the characteristics of misreading gestures,this paper extracts the eigenvalues from the whole connection layer to the convolution layer one by one,and visualizes them,and compares and analyzes the difference between the correct and wrong identification of similar gestures from the eigenvalue.In the course of the experiment,an important feature of distinguishing the mistaken gesture,the peak of the three-dimensional surface,is proposed for the first time.From a big data perspective,the peak of the surface corresponding to the same gesture on the same channel always tends to a fixed area,and the areas corresponding to the different gesture types are different.Based on this law,the characteristics of mistaken discovery and automatic error correction algorithm are put forward.Experimental results show that the recognitio n rate of this algorithm is about 15% higher than that of traditional convolution neural network algorithm.