以菲涅尔区理论为指导的基于低成本商用无线设备的人体形态识别研究

Because the optical target recognition based on video and image has the limitation of strict requirements on light and monitoring angle and high deployment cost, target sensing and recognition based on wireless signals have great application potential. However, the fine-grained wireless signal human perception recognition faces the challenge of relying on special equipment, high implementation cost, and perception accuracy influenced by the environment. This project aims to use the Fresnel zone theory to realize human body shape recognition using low-cost commercial wireless devices such as RFID. Research on the multi-path solution using beamforming and transmission signal extraction to improve perception precision and robustness; Research on the deployment of label deployment and reduction of blind angles based on the RFID signal Fresnel zone theory. Cost of implementation; Study of human perception recognition prototype systems using pervasive wireless devices in a typical environment. The essence of the project is to explore the influence of human form on wireless signals. The goal is to break through the “dead-end” problem and multipath factors in traditional wireless sensory recognition, and seek effective solutions for low-cost, high-precision, and robust human perception. The results will provide valuable references for wireless sensory applications such as behavior tracking and gesture recognition, and will be of great significance in promoting the application of Internet of Things.

由于基于视频和图像等光学目标识别存在对光线和监测视角要求苛刻且部署成本高等局限，因此基于无线信号的目标感知与识别具有巨大应用潜力。然而，细粒度无线信号人体感知识别面临着依赖专用设备、实施成本高、感知精度受环境影响的挑战。本项目旨在采用菲涅尔区理论，利用RFID等低代价商用无线设备实现人体形态识别。研究利用波束形成和透射信号精确提取的多径效应解决方法，提升感知精度和鲁棒性；研究基于 RFID 信号菲涅尔区理论的人体形态识别中标签部署、减少死角的感知模型优化方法，降低实施成本；研究典型环境中利用普适无线设备的人体感知识别原型系统。项目的本质是探索人体形态对无线信号的影响规律，目标是突破传统无线感知识别中的“死角”问题和多径因素，寻求低成本、高精度和高鲁棒的人体感知有效方案。成果将为行为跟踪和手势识别等无线感知应用提供有价值的参考，对推动物联网的应用发展具有重要意义。

本项目提出了一系列基于RFID的低成本无线感知应用系统。（1）被动式人体目标感知识别系统，通过深度学习融合步态特征和人体形态特征实现对动态人体目标的识别，利用神经网络和迁移学习降低对训练数据收集的成本，提取多维生物信息特征对静态人体目标进行识别，避免了传统方法需要专用设备造成高昂成本的问题。（2）材质识别和成像系统，探索无线信号穿过不同材质时信号幅度和相位的变化规律，提出多径干扰抑制算法降低多径等噪声的影响，设计特征提取算法去除材质尺寸的影响，构建优化模型缝合不同方向下无线信号在材质中的传输距离对材质横截面进行成像。（3）低成本可配置的通用感知平台，分析不同应用需求下传感器阻抗特征和无线信号频率响应的映射关系，设计无源可嵌入式的贴片式标签，提出可重构感知设计理论框架，避免传统方法需要专业设计知识和测试调试带来的人力及资源开销。本项目的相关研究成果陆续发表在CCF推荐A类会议UbiComp，A类期刊TMC上，被B类会议IPSN接收,授权技术发明专利2项。本项目的研究对物联网在实际生活中的应用，如人体目标识别、材质识别等具有重要的推动意义。