Research Of Gesture Segmentation And Recognition Based On RGB-D Images

In the recent research on the recognition and segmentation of gesture images,the gesture segmentation method based on RGB images often a?ected by complex background.As a consequence it is low e ciency to segment.Besides,the algorithm of gesture recognition based on RGB images is di cult to recognize because lack of three-dimensional characteristics.In order to solve these problems,we propose a method based on the recognition and segmentation of REBD gesture images.RGB-D image refers to an image representation combining RGB image and depth image,which can better represent the color feature and three-dimensional feature of the scene.In order to get depth images,This article uses a binocular mobile phone to acquire binocular image and synthesizes depth images based on binocular vision principle.Then,we segment gesture area of the depth image and the RGB image,and remove background region using the threshold segmentation method and the k-means pixel clustering method.Finally,this paper uses the deep learning algorithm based on LeNet-5 neural network to recognize RGB-D images.In the design of experiments,this paper used 9 gestures representing numbers 1to 9,and collected a large number of RGB gesture images of di?erent subjects in different backgrounds.Then we synthesis depth images by preprocessing.Subsequently,gesture segmentation and recognition experiments were carried out using RGB images and RGB-D images respectively.The experimental results show that the gesture segmentation method based on RGB-D image can segment the gesture area accurately and e?ectively,and the interference of complex background image is avoided.The recognition accuracy of gesture algorithm based on RGB-D image is higher,reaching 92.54%.