机场行人状态智能无线感知方法研究

Airport person status sensing is a new technique which tries to sense information such as crowd distribution and person location, speed, activity, or identity. It could provide fundamental information for the airport. This proposal tries to sense the person status by analyzing their influence on surrounding wireless signals, and provides a novel new method to realize airport person status sensing without equipping the persons with any device. Specifically, this proposal investigates the influence models, feature extraction methods, and sensing algorithms for the emerging wireless sensing technique. Firstly, we investigate and reveal the influence models of a human body on wireless links. We propose the radio tomography based crowd distribution model, spectrum feature based velocity model, modal analysis based identity model, parameter projection based activity model, and shadowing effect based location model. Secondly, we propose a method to transform multi-domain wireless link measurements into radio images and explore methods to extract features based on the radio images. We propose a deep-learning networks based method and a time-spatial clustering analysis based method to extract deep features and spatial structural features from the radio images, respectively. Finally, we develop a robust high performance classification algorithm based on the collaborative representation and transfer learning techniques. We will also develop a prototype sensing system to evaluate the proposed methods. The aforementioned work would provide basic models and algorithms for the emerging airport person status intelligent sensing technique, and would provide fundamental support for the realization of the goal of smart airport and safe airport.

机场行人状态感知是一种获取人群分布以及人体位置、速率、动作、身份等综合信息的新技术，它可为机场管理提供基础的行人状态信息。本课题拟探索一种通过分析机场内行人对周边无线链路的影响模式实现行人状态智能感知的新思路，以期在行人不携带任何设备的条件下实现状态感知。本课题拟从模型、特征、算法、系统四方面展开研究。首先，揭示人群/人体目标对无线链路的影响机理，提出基于射频层析的人群分布模型、基于频谱特征的目标速率模型、基于模态分析的目标身份模型、基于参数映射的目标动作模型、以及基于遮蔽效应的目标位置模型。其次，探索通过构建多域射频图像提取链路特征的新思路，提出基于深度学习提取图像深度特征以及基于时空聚类提取图像结构特征的新方法。最后，研究基于协同表示及迁移学习的高效分类识别算法。本课题将设计原型实验系统验证上述方法，以期形成完善的机场行人状态智能无线感知新方法，为实现智慧机场与平安机场提供基础支撑。

本课题以机场行人状态感知为应用背景，探索了基于无线感知技术的行人状态感知新方法。本课题针对行人状态智能无线感知中的模型、特征、算法、系统等方面展开研究。在模型方面，提出了基于多维复用策略的多目标无线感知模型与基于多普勒本征谱的目标运动模型；在特征方面，探索了无线信号健壮深度特征提取方法、无线手势健壮特征构建方法、基于自适应对抗学习网络的无线信号特征提取方法；在算法方面，设计了智能无线感知算法、小样本条件下基于元学习的无线感知算法、基于生成对抗网络的无线感知虚拟样本构建算法；在系统方面，研究了基于网络剪枝量化的轻量化无线感知系统设计方法与基于空间分集的多人无线手势识别系统设计方法。上述研究成果发表SCI论文10篇，其中9篇为IEEE期刊论文。论文发表在IEEE Wireless Communications、IEEE Transactions on Vehicular Technology、IEEE Internet of Things Journal、IEEE Sensors Journal、China Communications等期刊，申请发明专利3项，同时，成果在企业得到应用推广。项目负责人作为第二完成人获得2020年辽宁省自然科学奖二等奖。本课题系统地探索了人体状态无线感知新方法，有效地促进了该领域的理论发展，为其赋能机场行人状态感知等智能感知应用提供了方法支撑。