Research On Static Gesture Recognition Algorithm Based On Neural Network

Gestures are the most natural and intuitive way of communication in people's daily life.Gestures contain rich information,and gestures can express some natural language and written language hard to convey meanings.At present,the equipment used in human-computer interaction has been greatly improved in both performance and calculating ability,so the research on human-computer interaction technology based on gesture has become a hot research topic.Static gesture can be regarded as a transient state of dynamic gesture,so the research on static gesture recognition has important significance.The research is helpful for dynamic gesture recognition in the future.This paper focuses on the static hand gesture recognition technology to do a detailed study,and the main research contents are as follows:(1)According to interference caused by noise to the collected gesture image,firstly study the denoising principle of median and mean filtering,and the experimental results show that the two algorithms will cause the image unclear and make the edge lost.In order to solve the above problems,this paper adopts an adaptive median filtering +FNLM denoising algorithm,which is proved to be able to protect the image edge and so on.(2)For hand gesture segmentation,deeply study the two methods based on skin color and edge.The experiment found that RGB and HSV color space for skin color clustering effect are not very good,and they will segment the background color as the skin color.However,YCbCr space is better than the former two kinds of color spaces,so it is more suitable for skin color segmentation.And through study for several edge detection operators,the Canny operator can find accurate position and false edges are few.The disadvantage of above two methods is that it is easy to segment the background area as a gesture region.This paper combines YCbCr color space and Canny operator to implement gesture segmentation.The experimental results showed that the methods can divide a perfect gesture from the background and can also solve the phenomenon of false detection.(3)Study the feature extraction method of gesture based on SIFT,and it was found that this method has high complexity,long recognition time and other shortcomings,so this paper uses SIFT based on PCA feature extraction dimensionality reduction algorithm,to reduce the dimensionality of the feature descriptors extracted by PCA,which can reduce the complexity and improve the recognition rate.(4)For gesture recognition,this paper uses the BP neural network,and designed gesture recognition process based on BP neural network of SIFT of PCA dimensionality reduction.For different hidden layer nodes and learning rate,study their relationship with the recognition rate,and find suitable structure of BP neural network for this paper.Through the statistics analysis of error identification and contrast with other literatures,find that recognition method proposed in this paper can solve the influences of illumination,rotation and other factors on gesture recognition,and improve the recognition rate.