Research And Application Of Gesture Recognition Method Based On Kinect

With the rapid development of computer science,the field of human-computer interaction is becoming more and more diversified.It is no longer simply limited to the traditional keyboard,mouse and other input devices.Gesture recognition is a hot topic of natural interaction,because its intuitive,flexible,rich expression and other characteristics are more in line with people's habits in daily life.The use of visual gesture recognition method has become the mainstream method in this field,through the acquisition and processing of gesture images,the final output of the ideal result.Using the second generation Kinect sensor as the acquisition device,not only color information can be collected,but also depth information and bone information can be obtained.In this paper,static gesture recognition and dynamic gesture recognition methods are studied and improved respectively,and the improved algorithm is applied to the motion-sensing game developed on Unity3 D platform for practice.The main work contents are as follows.(1)Introduce the working principle of Kinect sensor,etc.,and add the judgment of 0value point,that is,the depth pixel point that cannot be obtained,on the basis of the traditional median filtering algorithm for the extracted depth image,so as to improve the original algorithm based on the 5*5 window for median filtering and noise reduction,and improve the efficiency.The common gesture segmentation algorithms,such as neighborhood segmentation based on depth image and color template matching,are experimented,and the shortcomings of each algorithm are analyzed.Finally,an algorithm combining depth information neighborhood method and YCB 'CR' color template matching method is proposed to segment the target gesture from the image.(2)the Border-Following algorithm is introduced in detail,and applied to gesture to the edge of the image contour extraction,the K-curvature algorithm and convex hull algorithm aiming at the condition of the fingers together on the basis of a "refers to the wide correlation method used at the same time,based on" palm finger test as an aid to improve the accuracy of finger detection,The static gesture recognition experiment was carried out with decision tree model based on the four feature parameters: the number of effective convex defects,the feature Angle of effective convex defects,the ratio of the contour area to the area of the outer circle and the number of fingers.(3)In the dynamic gesture recognition method,in order to enhance the real-time performance,the Angle feature of upper limb joint is used as the feature vector for matching recognition Dynamic Time Warping(DTW)algorithm is applied the classical DTW algorithm Two improved methods are proposed.One is the slope of the DTW algorithm neat paths limit within 0.5-2,at the same time limit the search scope based on the quadrilateral in the range,using look-up table method to replace every time the edge of the repeated calculation,improve the efficiency,the second is based on traditional DTW algorithm to optimize the parameters of the way to make two templates neat paths more approximation diagonal,thus improve the accuracy.(4)Based on the typing parkour game on the Unity3 D development platform,the improved dynamic gesture recognition method is used and applied in the game to increase the motion-sensory interaction module.From the experimental results,it can be seen that the accuracy is relatively ideal and the playability of the game is increased at the same time.