Boosting RFID Based Passive Sensing In Smart Environment

Smart environments can sense human behaviors and ambient conditions with various devices for accurate,efficient and intelligence services.Passive sensing which harvests energy from the environment for computing,sensing and communication,can reuse existing communication infrastructures.Nowadays passive sensing has become the foundation of smart environments.As one of the most representative passive sensing technology,Radio Frequency Identification(RFID)is not only an automatic identification technology,but also a sensorless sensing technology.The RFID backscatter signal can be used for localization,activity recognition and vital sign sensing,which has shown its great potential in the field of passive sensing.Although RFID based sensing has drawn extensive attention from both academy and industry,many challenges still exist when RFID based passive sensing technology is applied in real scenarios.In the field of RFID based localization,existing systems with high accuracy like synthetic aperture RFID localization systems usually ensure positioning accuracy depending on the spatial diversity of noise.However,they usually suffer a lot from the spatial ambiguity due to the low reading rate of a single tag in noisy environments,while the ubiquitous aperture position error will lead to the degradation of the localization performance.In the field of RFID based activity recognition,existing works focus on action or activity recognition in a specific scenario,and recognition accuracy usually depends on precise segmentation and other preprocessing methods.As a result,the recognition of an activity begins only after this activity is finished,which cannot meet the requirement of real-time applications.In the field of RFID based vital sign sensing,small signals caused by respiration and heartbeat are easily overwhelmed by environmental noise and intense motions,while difficulties may arise from the long wavelength and the discontinuity of RFID phase data flow caused by frequency hopping.In response to these challenges,this dissertation constructs the study from four respects as follows:· This dissertation discusses the theoretical and practical basis of RFID based sensing.Based on the RFID communication model,this dissertation explores the relationship between the low-level signal characters(phase information and received signal strength)and the tag location and the nearby human activity from different perspectives on multipath signal components of the received signal,including ignoring,suppressing,measuring and extracting these components.Then this dissertation also studies these low level signal characters in practical applications and influences of RFID devices,which reveals the practical basis of RFID based sensing.· This dissertation proposes two RFID based localization methods.After analyzing the RFID signal model,this dissertation transforms the localization problem into a sparse signal reconstruction problem,and establishes a compressed sensing based sparse signal reconstruction method for localization.It improves the localization accuracy with phase calibration based on the frequency and angle of arrival response,and achieves real-time tracking with a particle filter.In addition,this dissertation takes the aperture position error as a part of the localization model to combine the aperture position error compensation with the sparse signal reconstruction.It estimates alternately the target location and the aperture position error with an iterative algorithm using a single antenna.Extensive experimental results show that these two methods can achieve centimeter-level accuracy in the presence of low tag reading rate or aperture position error.· This dissertation proposes several RFID based activity recognition methods.It introduces a method to encode the phase flow based on vector quantization encoder and extract the inner temporal relationship of activities,without the demand of pre-segmentation and predefined key actions.Based on the code stream from the encoder,this dissertation leverages HMM for the outside temporal relationship between activities to recognize workflow steps and other complex activities.Due to their disadvantages on real-time performance and generalization,this dissertation proposes an RFID based activity recognition framework with deep learning,to directly extract both inner and outside temporal and spatial features of activities.This framework achieves negative delay activity recognition with support vector machine classifier,and obtains excellent generalization with adversarial learning.Extensive experimental results show that these methods can achieve high recognition accuracy in different environments and different applications such as Tag-on-Body or Device-Free gesture recognition and workflow recognition.· This dissertation proposes a contactless respiration and heartbeat monitoring method and leverages a sequence of signal processing on temporal phase information from the tag array near or on body to smooth phase data flow.After intense motion detection,this dissertation designs a signal separation approach based on tag empirical mode decomposition(tag-EMD)to obtain fine-grained respiration rate and heart rate.Furthermore,the monitoring method can also detect abnormal respiration.Extensive experimental results in tag-on-body and device-free scenarios validate its wide applicability and high reliability for fine-grained respiration and heartbeat monitoring.Firstly,this dissertation summarizes existing works of the passive RFID sensing technology,including the research progress and limitations.Secondly,this dissertation discusses the theoretical and practical basis of RFID based sensing.Then this dissertation proposes two RFID based localization methods with sparse measurements and aperture position error compensation and explores the real-time activity recognition method as well as designs an RFID based contactless respiration and heartbeat monitoring method.At last,the conclusion is drawn and future research is prospected.