Number Gesture Recognition Based On Depth Data

Nowadays,computer has become an essential part of human society.The mouse,keyboard and other hardware equipment have used for operating the computer at the beginning.With the depth and development of Human natural interaction(HCNI)technology,the market demand is quietly changing,and the directly,naturally,friendly communications with computer have been more and more wanted out of the bondage of hardware.Gesture as one most important ways in HCNI,can not only express the rich amount of information,but also has the advantages of intuitive,nature and efficiency,and the other posture often cannot be compared;As a result,gesture recognition technology has become a hotspot based on image in recent years.Because the gesture system is too large.From the foundation first,the recognition of ten kinds of Number Gestures in Static Gestures have completed by extracting all apparent characteristics in hands' depth image.Firstly,the experimenters' image data stream obtained by using OptiTrackTM?DPC640/30 Stereo Vision System;Such as Grabcut,the color image segmentation methods have likely under-segmentation due to influence of similar skin color and environmental light;In addition,using the fixed threshold for segmentation,it may make the segmentation posture is too stiff and the Human-Computer Interaction is unnatural in the process;For the sake of those problems,using depth image,combined with skeleton tracking technology based on machine learning,real-time tracking of the experimenter hand bone invasion have realized,and the accurate segmentation of the hand image has completed by using the appropriate dynamic threshold.Secondly,an operation of the repair and binary for segmenting hand image have completed.On the one hand,by adopting the method of combination of morphological and median filtering,a lot of noise in hand's area eliminated and hand's image contours smoothed;On the other hand,the improved Algorithm of Otsu is used to determine the threshold,it realized the hand's image binarization and simplified the expression of image based on retaining the original characteristics,which is convenient for the following operation.Thirdly,the extraction of all hand's apparent characteristics have accomplished.All contours of the hand's image detected by Canny Operator,the judgment of the inner contour and outer contour tracking achieved by Suzuki85 Algorithm;Centroid of hand was determined based on distance transform;Comparing the Method of Convex Hull and Curvature,the numbers of finger tips were determined by the method of distance combining with outer contour and centroid feature;The bending characteristics of finger judged by the method of left-neighborhood,and the angle between two finger tips and centroid determined according to the Cosine Theorem.Finally,the decision basis and identification procedure of ten kinds'Number Gestures have given.In the Windows 7(64 bit)Operating Environment,through the Microsoft Visual Studio 2013 platform based on DoNET Framework 4.5,with the Kinect for Windows Drivers,EmguCV 3.0 Toolkit and so on,the ten kinds of Number Gestures'recognition have realized by using XAML and C#Language on WPF and Winform.