面向多信息融合的RFID定位与活动识别方法研究

As the key enabling technology of Internet of Things (IoTs), radio frequency identification (RFID) has great importance to the development of our country’s economy, and it has attracted much research attention in recent years. RFID has the potential to provide pervasive and low-cost solutions to object localization and users’ activity recognition, and thus can be widely used in many fields such as smart logistics, smart medical, and elderly healthcare. Existing works on RFID localization and activity recognition use only single type of information, and cannot achieve high localization accuracy, real time, and high scalability simultaneously. ...In this project, we study RFID-based localization and activity recognition from a multi-type information fusion angle, with the objective to provide highly accurate, real-time, and scalable solutions. The main research tasks include the following subtasks. 1) We need to design efficient mechanism to fuse multiple type information to grab their advantages and avoid disadvantages of different type of information, with the objective to improve localization accuracy as well as reduce localization latency. 2) For simultaneous localization of a massive number of tags, we find a new problem called sparse reading of single tag. The problem is that for each single tag, only a few number of readings can be obtained in a given time period, which causes the readings of single tag sparse. For this new problem, we propose several data regression-based methods to compensate the sparse readings, with which we can support simultaneous localization of multiple tags with high accuracy. 3) Based on the movement trajectory of RFID tags, we study RFID based activity recognition with machine learning methods. Moreover, we study the recognition of a set of associated activities and a series of activities with the hidden Markov model (HMM), which are both new problems that have not been studied in previous researches. 4) We study the deployment and schedule of readers and antennas from the perspective of RFID localization and activity recognition. More specifically, we study the deployment and schedule of multiple readers (antennas) with objectives not only to guarantee coverage, but also to guarantee fair reading of different tags (i.e., different tags should be interrogated nearly the same number of times in one second). This type of deployment and scheduling problems are totally different from previous similar problems and need new solutions. ..The researches in this proposal provide a novel angle to study RFID-based localization and activity recognition. We find several new problems that didn’t rise in previous researches, and provide novel solutions to them. The output of this proposal can provide holistic solution to highly accurate, real-time and highly scalable RFID-based localization and activity recognition, which provides ways to implement pervasive smart sensing in IoTs. The principle investigator has very good background in RFID systems and wireless localization systems, which can guarantee that the proposal can be finished successfully.

无线射频识别(RFID)作为物联网最核心的技术之一，其相关研究对国家经济民生非常重要。RFID可以提供普适、低成本的定位和活动识别方法，在智慧物流、智能医疗、老年人健康护理等领域有广阔的应用前景。已有工作基于单类信息，无法兼顾高精度、实时性和高扩展性。本课题从多信息融合的新角度研究RFID定位和活动识别，包括：1) 设计高效的多信息融合机制来提高定位精度并降低时延；2）针对多标签并发定位中的读取数据稀疏这一新问题，提出基于数据补偿的解决方法；3）研究基于标签运动轨迹和移动性检测的用户活动识别，在此基础上利用隐马尔科夫模型研究关联活动识别和活动序列识别等新问题；4）从与定位和活动识别联合优化的角度研究阅读器和天线的优化部署调度。本课题从一个新的角度对RFID定位和活动识别展开研究，针对若干新问题提出了可行的解决方案，可望实现高精度、低延时、可扩展的目标，对于实现物联网普适智能感知有重要意义。

针对高精度、低延时和可扩展的目标，本项目主要从多信息融合的角度研究RFID定位和活动识别方法，取得了如下进展。在RFID定位方面：1）提出了多信息融合的RFID标签定位算法，通过融合信号强度信息和相位信息，实现了对标签的厘米级绝对定位，将定位时延降低到秒级以下。2）针对标签相对定位，提出了一种基于相位变化率的高精度低时延相对定位方法，实现了标签密集部署时的高精度定位，在标签间距1厘米时相对定位精度达到98.6%并将时延降低到0.8秒。3）提出了利用跳频信息的视距/非视距RFID信号判定方法，提取信号强度和相位值的相关统计量作为特征，利用随机森林方法进行视距/非视距信号判定，检测精度达到0.98，检测时延低于0.4秒。在RFID活动识别方面：1）提出了一种结合信号时域特征和频域特征的实时活动识别方法，在18个手势上达到了0.97的识别精度，并且将识别时延降低到100毫秒以下。2）提出了一种基于RFID的面部微动作检测识别方法。该方法考虑了不同口型动作引起的嘴部多路径效应，在包含20个中英文单词的词汇表中达到了88%的识别准确率。3）提出了基于配对RFID标签的移动用户生命体征监测方法，通过监测两个具有固定几何关系的标签来消除移动性的影响，在商用RFID设备上呼吸率监测误差小于0.54 BPM (Breath per minute)，呼吸时长监测误差小于10%。4）针对多标签下的稀疏数据补偿机制展开了研究。提出了一种基于标签间相位插值的统计特征来对目标用户进行认证的机制。在5个用户的情况下可以有效对采用同一手势进行认证的不同用户进行区分，区分精度达到92.8%。在阅读器调度方面：1）提出了可以同时对多个用户的搜索请求进行并行处理的多用户标签搜索协议PTS(Parallel Tag Searching)，相比传统方法搜索效率在同时服务10个用户时可以提升7.54倍。2）提出了两种阅读器负载均衡协调算法，在标签分布均匀和分布不均匀时可有效提升吞吐量近1倍。发表（录用）相关论文44篇，其中CCF A类论文10篇，申请发明专利8项（授权4项）。协助培养博士生3名，培养硕士生9名。