基于级联卷积神经网络的彩色图像三维手势估计  

Color Image 3D Gesture Estimation Based on Cascade Convolution Neural Network

作　　者：刘玮 戴仕明[2] 杨文姬[2] 杨红云[2,3] 钱文彬 LIU Wei;DAI Shi-ming;YANG Wen-ji;YANG Hong-yun;QIAN Wen-bin(School of Computer and Information Engineering,Jiangxi Agricultural University,Nanchang 330045,China;School of Software,Jiangxi Agricultural University,Nanchang 330045,China;Key Laboratory of Agricultural Information Technology of Colleges and Universities in Jiangxi Province,Nanchang 330045,China)

机构地区：[1]江西农业大学计算机与信息工程学院,南昌330045 [2]江西农业大学软件学院,南昌330045 [3]江西省高等学校农业信息技术重点实验室,南昌330045

出　　处：《小型微型计算机系统》2020年第3期558-563,共6页Journal of Chinese Computer Systems

基　　金：国家自然科学基金项目(61462038,61562039,61502213)资助.

摘　　要：估计手的三维姿态是人机交互中重要的组成部分.针对从单个彩色图像估计准确的三维手势困难这一问题,提出了一种基于级联卷积神经网络的估计方法,该级联网络分三阶段,手部掩膜估计、二维手势估计和三维手势估计,三阶段级联网络进行端到端的训练,可以实现相互促进,最终优化三维手势估计的准确性.在两个公共数据集上进行了实验,实验结果表明该级联网络产生了卓越的三维手势估计精度,验证了该级联网络的有效性.It is an important part of human-computer interaction to estimate the 3 d gesture of hand. For the difficulty in estimating the accurate 3 d gesture from a single color image,this paper proposes an estimation method based on cascade convolution neural network.The cascade network is composed of three stages,hand mask estimation,2 d gesture estimation and 3 d gesture estimation and the three stages conduct end-to-end training which can realize the mutual promotion,eventually optimizing 3 d gesture estimation accuracy. Experiments are carried out on two common datasets. The experimental results show that the cascading network produces excellent 3 d gesture estimation accuracy and verifies the effectiveness of the cascading network.

关 键 词：级联卷积神经网络 手势估计 三维手姿态 彩色图像 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]