Monocular Gesture Recognition And Its Application In Android Platform

Recently,as the increasing popularity of human-computer interaction technology,it has been widely used in various fields such as scientific research,life and production,.etc.Gesture recognition technology based on monocular vision is one of the most important parts of human-computer interaction technology.It involves several key technologies such as gesture segmentation,feature extraction,tracking,gesture recognition and so on.In actual application,some factors such as illumination changes and complex background interference can affect the performances of gesture recognition algorithms easily.In addition,considering the requirements of real-time interactivity,the complexity of the algorithm is limited.These make the dynamic gesture recognition technology based on monocular vision be a difficult job in the field of human-computer interaction.This paper presents an improved gesture recognition algorithm with low computational load.And by studying the overall architecture of Android,gesture recognition is applied to the interaction of the Android platform to verify its validity and practicality.The research of monocular gesture recognition technology on Android platform includes the following aspects:In terms of gesture segmentation,the paper deeply studies the principles,advantages and disadvantages of the skin color segmentation method and inter-frame difference method.In this paper,an adaptive hand identification image segmentation method based on color histogram is applied.The method reduces the noise that generated by the surrounding environment and lighting by acquiring dynamic color information.Compared with the traditional color segmentation,the proposed method has less information interference when extracting the hand image.In terms of gesture features extraction,the paper analyzes the existing hand contour feature extraction methods.A fingertip-fingers hand contour recognition method is proposed to solve the problem that currenthand contour feature extraction methods are ineffective in a complex background.Using the curvature characteristics of the top of finger and the bottom of the finger accurately delineate the hand range into the image which been smoothed and extracted image contours.The method can identify a variety of gestures in the complex background.In terms of the dynamic gesture tracking,the paper studies the existing gesture tracking technology and analyzes the Camshift gesture tracking algorithm principle,realization and existing problems.We propose an improved algorithm of Camshift based on the historical movement matrix.The method combines with the adaptive extraction of hand to improve the dynamic gesture tracking accuracy in complex background.Experimental results show that,in the hands and face in the dynamic overlap case,the success rate of the proposed algorithm increased by 80%compared to the classic Camshift algorithm.In terms of the dynamic gesture recognition,through the abstract expression of gestures,the analysis of gestures and gestures vector definition,a dynamic gesture recognition method based on time-axis compression direction vector is proposed.The vector of locus is attenuated in the direction of time axis.The method which records the time axis direction vector to judge the dynamic gesture trajectory and recognize the dynamic gesture,has a small computational load,and suits for practical applications.In terms of Android platform realization,the paper deeply studies the Android platform architecture,JNI technology,V412 framework under Linux and Camera architecture under Android.The paper implements the camera driver under Android,OpenCV migration and dynamic gesture recognition algorithm under Android,and completes the Android application demo showing that identifying six kinds of gestures.