摘要
基于深度学习的视觉手势估计一直是计算机视觉领域的重点研究课题之一,随着深度学习和神经网络相关研究取得了巨大进步,针对手势估计中的高自由度、肤色、环境干扰、遮挡等问题已经远远优于传统方法。基于深度学习的三维手势估计主要是通过构建神经网络,对图像特征进行抽象化分析和理解,从而预测出手指关键点的三维坐标以及角度等信息,进而构建出手掌模型。准确的三维手势估计可以快速推动AR/VR行业的发展,因为沉浸与交互是AR/VR的关键要素,通过视觉手势交互可以为用户提供更方便、快捷、逼真的AR/VR互动体验。本文首先对当前手势估计方案进行阐述,了解到手势估计各方案的优缺点,然后介绍了基于深度学习的手势估计方法、相关数据集和评价指标,最后根据各研究结果,对当前三维手势估计所面临的挑战以及未来发展进行阐述。
analysis and understanding of image features,so as to predict the three-dimensional coordinates and angles of key points of fingers and then build a palm model.Accurate 3D gesture estimation can rapidly promote the development of AR/VR industry,because immersion and interaction are the key elements of AR/VR.Through visual gesture interaction,users can provide more convenient,fast and realistic AR/VR interactive experience.In this paper,firstly,the current gesture estimation schemes are described,and the advantages and disadvantages of each gesture estimation scheme are understood.Then,the gesture estimation method based on deep learning is introduced,and the related data sets and evaluation indexes are introduced.Finally,the challenges and future development of current 3D gesture estimation are described according to the research results.
作者
武胜
秦浩东
WU Sheng;QIN Haodong(China Nanhu Academy of Electronics and Information Technology,Jiaxing Zhejiang 314001,China)
出处
《智能计算机与应用》
2023年第11期232-238,共7页
Intelligent Computer and Applications
关键词
手势估计
深度学习
关键点检测
神经网络
gesture estimation
deep learning
key point detection
neural network
