应用于FPV眼镜交互系统中的手势识别研究  

FPV-GLASS interaction system based on 3D gesture recognition

作　　者：吴伟[1] 肖中银[1] Wu Wei;Xiao Zhongyin(School of E1ectronic and Information I~Tngineering,Shanghai University,Shanghai 201900,China)

机构地区：[1]上海大学电子信息工程学院

出　　处：《电子测量技术》2018年第16期85-90,共6页Electronic Measurement Technology

摘　　要：FPV眼镜是一种常用于旋翼无人机、固定翼无人机的图像显示和控制设备。为解决因传统交互方式造成的使用场景单一等问题,提出一种应用于FPV眼镜的手势识别交互系统,并重点研究其中的手势识别技术。系统通过对卷积神经网络进行优化设计,并对训练中的超参数进行精细调节,使系统在静态手势和连续动态手势的实际测试中均获得了超过80%的准确率。同时,为实现在嵌入式系统中使用该技术,使用网络优化、并行化处理和Staple追踪器的方法使系统达到实时性要求,具有很高的实用性。FPV is a kind of image display and control equipment commonly used in drones and fixed wing uav. In order to solve the problem of single use scene caused by traditional interaction method, the paper proposes a gesture recognition interaction system applied to FPV glasses, and mainly focus on the gesture recognition technology. The system based on CNN optimization design, the super parameters and the training with fine tuning makes the system in the actual static gesture and dynamic gesture real test have won more than 80 ~ accuracy. At the same time, in order to achieve using the technology in embedded system, this paper use network optimization, parallel processing and Staple tracking technology makes the recognition efficiency reached the requirements of real time, the system has the very high value.

关 键 词：FPV眼镜 手势识别 Staple跟踪器 卷积神经网络 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术] TN06[自动化与计算机技术—计算机科学与技术]