Gesture Position Detection Based on Generative Adversarial Networks

Abstract

In order to improve the real-time and accuracy of gesture position information acquisition in gesture interaction, this paper proposes a gesture position detection method based on generative adversarial networks. Firstly, this paper collects gesture images from two perspectives by binocular camera, and then uses semi-supervised generative adversarial network to segment gesture images. This method overcomes the problem that traditional semantic segmentation ignores the correlation between pixels, and obtains better gesture segmentation images. Secondly, fingertip detection and three-dimensional reconstruction of fingertips are performed on gesture semantic segmentation images. Finally, the method is verified by experiments, and the gesture segmentation results and fingertip detection results are evaluated by using mIoU value and fingertip detection accuracy. The experimental results show that the mIoU value of gesture segmentation can reach 94.8, and the accuracy of fingertip detection can reach 96.5.