The Research And Application Of Gesture Recognitionbased On Kinect

With the advancement of virtual reality technology and the development of humancomputer interaction equipment,the interaction mode is gradually diversified.Among them,gesture has become the preferred object for many scholars because of its intuitive,natural and informative characteristics.In order to meet the application requirements,gesture recognition technology needs further research and improvement in accuracy and environmental adaptability.This paper mainly focuses on the research and improvement of fingertip detection and gesture recognition algorithm,and implements a projection interactive system to demonstrate the feasibility and recognition effect of this scheme.The main research work is as follows:(1)The traditional gesture region segmentation and tracking methods are studied.By analyzing the characteristics of the multi-channel image and comparing the results of the segmentation,the Kinect device is finally selected to acquire the data.After the image preprocessing,the gesture region is segmented by combining the skin color feature and the depth information,and the continuous segmentation operation is performed to optimize the segmentation result;On the tracking filter optimization algorithm,the results of CA and CV model tracking target experiments are analyzed.Finally,CA model filtering is used to track the prediction gestures to achieve higher response accuracy.(2)An improved multi-fingertip detection method is proposed.The shortcomings of fingertip detection algorithms based on K-curvature and convex hull are analyzed respectively.The combination of the two methods is used to identify and detect multiple fingertips.Experiments show that K-curvature algorithm based on convex hull can effectively detect and locate fingertips.Furthermore,the T and K parameters are analyzed for optimization.(3)An improved SVM gesture recognition method is proposed.Through analysis,it is concluded that a single feature is unstable in a complex environment.Therefore,the three characteristics of tandem fusion structure feature,Hu invariant moment and HOG are used,and PCA is used for dimensionality reduction,which is used as the final classification feature input SVM for identification.The identification method showed high accuracy and good environmental stability in the experiment.At the same time,this paper designs and implements a projection interactive system,and collects sample sets of 500 complete motion data of 5 experimenters for 8 gestures in the text,and further tests the algorithm model in the text.The experimental results show that the proposed scheme improves the recognition accuracy from 90.4% to 95.3%,and the accuracy is maintained above 93.7% in complex environments with poor illumination conditions,complex motion background and angular interference,and it has good robustness.Although this paper improves the accuracy and environmental adaptability of traditional gesture recognition methods,there are still some problems in depth image denoising and mouse jitter processing.Next,a lot of research and learning are needed to further optimize.