Research And Application Of Static Gesture Recognition Based On Monocular Vision

With the rapid arrival of the intelligent era,human-computer interaction technology has penetrated into people's daily production and life.As an important part of human-computer interaction technology,gesture recognition still has problems such as illumination changes affecting image quality,complex background interference gesture area acquisition,camera itself adding noise to the image and the application field of the technology is not clear.Aiming at the applicability and popularization of this technology,the algorithms involved in gesture image segmentation,gesture feature extraction and gesture recognition are studied and improved.The algorithm can not only overcome the surrounding complex environmental impact,but also make the recognition accuracy significantly improved.Finally,relying on the improved algorithm to design a smart home gesture recognition system.The main work and innovations are as follows:(1)In the gesture region segmentation,in order to solve the problem that the gesture region extraction is susceptible to interference with the skin-like background and light changes in the real environment.This paper compares the effects of RGB,HSV and YCbCr on the same skin color,and briefly summarizes the use of inter-frame difference algorithm and background difference algorithm.Innovatively,a gesture region segmentation algorithm using YCbCr color space and background difference method is used.The skin color segmentation algorithm,the background difference algorithm and the combination algorithm are used to segment the gestures in a more complex environment.Through experimental comparison,the combined algorithm can overcome the influence of the environment and extract more accurate gesture regions.(2)In the process of gesture feature extraction,in order to solve the problem that the single feature cannot accurately describe the gesture information and the combination of multiple gesture features is complicated.Firstly,an innovative research is designed to detect the number of fingertips based on the combination of curvature and fingertip circle characteristics.Compared with the traditional fingertip detection algorithm,the detection efficiency and accuracy of the proposed detection algorithm are higher.When studying and extracting the Hu invariant moment feature of the gesture binary image,after analyzing the test data,it is found that the first four lower moments of the 7 moments of the Hu invariant moment can better describe the information of the gesture image,and the last three High-order moments not only have long detection times,but also easily contain interference factors such as noise.In order not to introduce interference factors,the first four low-order moments are selected to describe the feature information,and the experimentally obtained data better verify that the Hu invariant moment has the characteristics of translation,rotation and scale invariance.The Fourier descriptor of the binary gesture profile is obtained experimentally.In order to verify the usability and accuracy of the Fourier descriptor,theFourier descriptors of different numbers are used to reshape the contours with the help of the inverse Fourier transform.experiment.Through the comparative analysis of the experimental results,it is confirmed that the top 10 Fourier descriptors can better describe the gesture contour without causing edge noise.Therefore,the three types of fingertips,Hu invariant moments and Fourier descriptors are simply linearly combined into a 15-dimensional data matrix to describe the static gestures.(3)In the process of gesture recognition and application,after comparing and analyzing several types of classifiers,BP neural network is used to complete the training and recognition of feature combination data.In order to determine the better neural network structural parameters,the recognition result is obtained by changing the number of hidden layer nodes and the learning rate.Then,the structural parameters are determined by comparative analysis of the recognition results,and the static gesture recognition has been studied.The algorithm carried out a comparative experiment,and it can be clearly analyzed from the test data that the recognition rate of the designed gesture recognition algorithm is significantly higher.And in the practical application,the smart home module circuit is simply built,and the gesture recognition system is better transplanted into the smart home system.The gesture recognition algorithm proposed in this paper has been verified in the design of smart home system,which can accurately complete the simple control of all kinds of electrical appliances in smart home,and has good practicability and popularization.