Integration Of Multiple Features And Compressed Sensing Gesture Recognition

With the rapid development of computer science and technology, communicationbetween people and computers more and more on operating experience have becomeincreasingly demanding, human-computer interaction from computer-centric and graduallymoved to the people, makes machines interaction more natural, harmonious, gesturerecognition has been an important research topic of human-computer interaction in virtualreality technology continues to develop today, gesture recognition as the basis for virtualreality technology has made great progress. However, based on visual gesture recognitiontechnology is still low there is still a recognition of the shortcomings of illumination, scaleand rotation transformation irritations.Based on computer vision gesture recognition process can be divided gesturesegmentation, feature extraction, and gesture recognition in three stages. In gesturesegmentation stage, we propose a method based on adaptive threshold YCbCr color space, themethod can effectively eliminate the influence of illumination, and can dynamically adjust thethreshold for different skin color segmentation users. Then based Hu moment handcomparison algorithm, from the divided color regions further gesture area defined; extractionstage characterized herein as extraction global features gesture (Hu Moments), and extractingthe local feature gesture (SURF), will feature fusion of two different weights according to theclassification used gestures, improved gesture recognition performance; in gesture recognitionstage, we propose a gesture classification based on compressed sensing, feature extraction testsample composed of ultra-complete redundancy dictionary, the characteristics of the testsample expressed as a linear combination of the corresponding sparse over-completedictionary redundancy, using L1norm for solving the optimization problem to achieve gestureclassification, the experimental results show that the method currently used widely andgesture recognition method compared, has a high competitive and characterized by theintegration of two shapes, illumination, scale, rotate more robust changes.