Research On Hand Gesture Segmentation And Recognition Based On Visual Attention Model In Complex Background

With the development of computer technology,human-computer interaction is very important in people's daily life.People began to pursue a more natural,smarter and more interactive way of interacting with human beings.Gesture-based information exchange is natural,convenient and can realize non-contact interaction,which can meet the new demands of human-computer interaction.Therefore,gesture recognition technology has been a hot topic in the field of human-computer interaction.At present,gesture recognition technology has made great progress,but there are still some problems in practical application.The main difficulties include the following two aspects:(1)How to eliminate the impact of complex backgrounds on gesture segmentation.In practice,the background often contains many complex shapes,various colors of non-gesture areas and the interference of human face and other body parts,and various changes in the color and brightness of light.In this environment,the method based on motion information can effectively eliminate the interference of the skin color regions,but it is difficult to extract the complete gestures.The skin color model-based method is difficult to adapt to the human skin color differences,light changes and can not exclude the skin color area.(2)How to extract effective gesture features.There are several degrees of freedom for people to make gestures,and there are differences in skin color and size between themselves.Geometries are relatively simple and less information can be extracted.Therefore,it is difficult to extract effective gesture features.Aiming at the existing problems of gesture segmentation and recognition methods,the main contributions and innovations of this thesis include the following:(1)In the aspect of gesture segmentation,the regional contrast model is applied to the gesture segmentation in view of the interference of the non-gesture area and the skin-like area in the background.A skin-region contrast model is proposed according to the deficiency of the model.In the model,Gaussian skin color model based on HSV and YCbCr color space is established,and the skin color saliency map is obtained,which is fused with the saliency map of regional contrast to obtain the final saliency map.Finally,a fixed threshold segmentation method is used to extract the gesture area and the experimental results show that the proposed method has better segmentation performance than the hand gesture segmentation method based on the regional contrast model.(2)In terms of gesture segmentation,in order to interfere with the complexion of human face and other parts of the body,a multi-feature Bayesian inference model is first adopted for gesture segmentation.This model extracts the shape and texture features through HMAX model,Exclude the interference of the skin color region;and quantify the color space of HSV and YCbCr to extract the color features to exclude the interference of other non-skin color regions.Finally,the obtained feature is taken as the input of Bayesian network to get the gesture spatial probability distribution map,which is the gesture saliency map.Experimental results show that this method effectively reduces the interference of complex background and improves the accuracy and recall of gesture segmentation.(3)In the feature extraction of gesture recognition,a fusion method of Hu moments and HOG features is proposed in this thesis.PCA dimensionality reduction and gesture recognition are implemented by using support vector machines.Finally,the experimental results show that this method has a higher recognition rate than single feature.