【作者】 杨磊；

【导师】 齐勇；

【作者基本信息】 西安交通大学 ， 计算机科学与技术， 2017， 博士

【摘要】 随着物联网技术的兴起,射频识别技术(Radio Frequency IDentification,RFID)越来越受到人们的重视,被视为本世纪最为潜力的技术之一。尽管RFID的发展如火如荼,甚至被预见为条码技术最有力的竞争对手和替代技术,但它现今还未能改变条码一统天下的局面,其原因在于还有很多基础的科学与技术问题尚未得到很好的解决。本文针对标签与阅读器的信号冲突,基于射频识别技术的防伪认证,以及标签移动察觉三个重要问题展开了深入的研究和分析。本文的主要创新点包括:第一、提出一组多阅读器环境下的冲突搁置与协同识别协议。信号冲突问题在RFID技术的诞生之日起就成为科研工作者竭力想解决的基础问题之一。当多于一个RFID终端(标签或阅读器)发射信号时,它们的信号在空中相互叠加,导致接收端无法将信号正确解析出来。根据发送端和接收端的不同,这些冲突又可细分为:标签冲突,阅读器冲突,和标签-阅读器冲突。由于RFID设备硬件计算能力有限,一般只能采用分时调度的策略规避这些冲突,所以相应的算法又被称为防冲突算法。本文针对上述三种信号冲突,提出了一组两阶段的协议栈。这些协议不仅能够有效避免阅读器冲突,还能协同相邻阅读器对公共覆盖的标签进行识别。其主要思想是根据是否发生阅读器冲突将目标标签划分为两类,一类是能够成功解析阅读器命令的"非争议性标签",一类则因为阅读器冲突而不能解析命令的"争议性标签"。本文发现80%的识别延迟主要来源于只占总标签数目20%的争议性标签,争议性标签是系统性能瓶颈的关键。本文首次提出搁置争议性标签的信号干扰,协同邻居阅读器识别的思想,区分对待上述两类标签。实验表明,与已有的工作相比,本文的方法能够提高6倍以上的识别吞吐量。第二、提出浑水问题及一种无识别批量标签认证方法。RFID标签因制造难度高、具有简单计算能力等特点,使其开始逐渐受到防伪领域的青睐,尤其是高端产品防伪,如药品、红酒和珠宝等。工业界采用序列号验证的方式进行防伪认证,这种方法虽然简单,但其安全性一直受到挑战。为了解决这个问题,相关的科研工作引入哈希函数,验证标签对序列号进行哈希操作后的签名。这种做法大大提高了标签的安全性,降低了标签被复制的风险。与传统方法一样,引入哈希机制后的算法仍然需要对标签进行逐个认证。当面对大批量标签时,这种逐个验证方式效率低下、不具扩展性、适用场景受限。本文提出了一种无识别的批量认证方法。该方法无需为每个标签建立独立会话时间片,即使发生信号冲突也能正确检测伪造标签的存在。该方法能够快速判断目标批次是否有效,即是否包含伪标签。在无效的情况下,能够结合组合分组测试方法和基于强度大小的空间划分,精确定位出这些假标签,填补机器与人类认知的沟壑。与以前的工作相比,可扩展性方面,MRBA的算法能够以高概率保证假标签在O(n·log(N/n))轮被检测出来;安全性方面,当探测的帧长达到最优化时,香农熵达到最大,系统被破解的概率趋近于零;精度方面,实验表明系统探测伪标签精度可达到99%以上,平均定位精度约为80%;效率方面,系统批量认证的时间仅仅占逐个认证方法所消耗时间的4%。第三、提出一种强渡流驱动的标签移动察觉方法。RFID技术作为自动识别技术的急先锋,最初目的为方便快速识别物体,取代手工录入,实现工业处理的自动化。但近年来,除了识别之外RFID的各种潜在应用也被逐步挖掘出来。例如,利用RFID技术对贵重物品进行实时监控,精确记录用户行为方式等。现有的基于RFID的监控技术只能给出"存在-或-不存在"的结果,这种粗放式的监控完全不能满足实时性和安全性要求较高的场景。本文通过理论分析和实验验证发现:标签的反射信号强度对所处位置非常敏感,可以利用这一规律准确并且高效的探测标签的移动。受到计算机视觉技术的启发,该算法将目标空间中的标签信号变化抽象为信号强度图,并提出采用基于混合高斯模型的背景相减法检测出移动标签。实验表明,这种强度流驱动的标签移动察觉技术能够获得运动监测92.34%的精度并且假阳性率维持在0.5%以下。更多还原

【Abstract】 Radio Frequency IDentification(RFID)has been widely deployed in a variety of applications and considered as one of the most promising technology in this century.This thesis performs a deep research on three issues:signal collisions among reader and tag,RFID based anti-counterfeiting,and motion perception.The main contributions of this thesis is given by:(1)The problem of signal collision emerged since RFID technology was invented,becoming one of fundamental issues in RFID area.When more than one client(reader or tag)emit their signals,the receiver cannot correctly resolve their IDs because these signals are overlapped together.According to the client types,these collisions are categorized into tag collision,reader collision,and reader-tag collision.Due to the hardware limitation of tag,the TDMA method is adapted in usual to avoid the collision.Correspondingly,these algorithms are called anti-tag collision,anti-reader collision,and anti-reader-tag collision.Aiming to resolve above three kinds of collisions,we propose a stack of protocols.These protocols are able to not only effectively avoid the signal collisions,but also significantly improve the identification rate and throughput.The basic idea is to divide the target tags into two types,contentious and non-contentious tags,based the influence taken form the reader collision.We find that the 80%delay mainly comes from the minor contentious tags which quantity occupies 20%of the total number.We first propose to shelve the contentious tags’ interferences,and jointly identify the non-contentious tags among neighbouring readers.The experiments show that our approach outperforms the throughput by 6 times compared with the existing algorithms.(2)RFID has received plenty of attentions in the area of anti-counterfeiting due to its capa-bility of computing and being hard to forged,especially for high value-added goods,such as jewelry,wine,pharmaceuticals and so on.In industry,a serial key is stored in the tag and used to help determine the tagged goods’ genuineness.Such method is simple but challenged in security.To address this issue,a hash function based method is introduced.However,above two methods are per-tag fashion based,which need perform the authentication tag by tag.We propose identification-free batch authentication for RFID tags.This approach works without need of per-tag authentication,even under tag collisions.In applying,our approach has several advantages:It can quickly determine whether there exist counterfeits in the batch,filter out the counterfeits through the technique of combined grouping test,and locate the counterfeits by space partition.Our experiments show that compared with the per-tag authentication,our approach significantly improve the authentication with regards to security,accuracy,effec-tiveness,and scalability.(3)As a pioneer in automatic technology,RIFD is motivated to automatically identify objects fast and conveniently.However,its potential applications have been widely studied in recent years,like securing valuable objects and mining custom’s behaviors.To the best of our knowl-edge,few works have been studied on the surveillance of tag motion.The existing methods output ’presence or absence’.Such coarse-grained results are hard to fulfill the requirements for real-time surveillance.We observe that the tag’s signal is hypersensitive to its position.We can leverage the change pattern of tag ’s strength to perceive its motion.Inspired by computer vision,we abstract the signal strength to a strength image in which these pixels mapped from the moved tag becomes to foreground.The Gaussian models of Mixture are used to detect the foreground pixels,which are then reversely mapped to the moved tags.Our evaluations show that the strength flow based perception of tag motion achieves 92.34%accuracy while maintain the false positive rate under 0.5%。更多还原