This article has been accepted for inclusion in a future issue of this journal.Content is final as presented,with the exception of pagination. XIE er al:MANAGING RFID DATA:CHALLENGES.OPPORTUNITIES AND SOLUTIONS 13 TABLE VII COMPARATIVE STUDY OF THE LOCALIZATION SOLUTIONS Accuracy Time-efficiency Deployment Cost Hardware Cost Range-based localization Low High Low Low using RSSI(RL) RSSI fingerprint-based lo- High High High Low calization (FL) Reference tag-based local- Median Low Median Low ization using RSSI(RTL) Phase-based localization High Median Low High (PL) Angle of Arrival based lo- High Median Low High calization (AAL) RFID faults greatly affect the localization accuracy,in order performance in terms of accuracy,time-efficiency,deploy- to tackle this problem,Zhu et al.propose an effective fault- ment cost and hardware cost.Here,the accuracy denotes tolerant RFID reader localization approach suitable for thethe precision of localization,the time-efficiency denotes the above situations,and illustrate how to measure the quality locating speed,the deployment cost denotes the human cost of a localization result [82].In order to deal with the noisy of deploying the localization system and getting the training RFID readings gathered in object tracking applications,Yang data,the hardware cost denotes the fixed cost of the RFID et al.propose a hybrid method for tracking mobile objects readers and tags in the localization systems. with high accuracy and low computational cost [83],such that Table VII depicts the performance comparison of the lo- the computational cost of filtering the noises can be greatly calization solutions.In order to make it clear and concise reduced. in the following we respectively use the abbreviations RL, FL,RTL,PL and AAL to denote the corresponding location D.Activity Sensing schemes.In regard to the accuracy,FL,PL and AAL all Activity sensing has drawn many attentions in recent years, achieve high accuracy in localization,the localization errors and it has yielded lots of research results.Most of them are can conventionally be limited to at most 0.5m.RTL achieves related to vision-based schemes and surveillance technologies. a median accuracy by limiting the errors to at most Im.RL The surveillance technologies are further classified into the achieves a low accuracy with the localization errors of at most Bluetooth,infrared,ultrasonic and sensor based solutions. 5m.In regard to the time-efficiency,both RL and FL achieve These solutions mainly require fairly expensive equipments high-efficiency since they only require to obtain the RSSI to capture scenarios as video or other sensor data,moreover, from the target tag.PL and AAL achieve median-efficiency a lot of energy is expended for these equipments during the as more refined data are required and more computations are procedure of activity sensing.These issues greatly limit the involved in the localization.RTL achieves low-efficiency in range of applications,as most pervasive applications cannot time,because the RSSIs from multiple tags are required to be afford such a huge overhead in terms of both deployment cost obtained,which can be very time consuming.In regard to the and energy consumption. deployment cost,RL,PL and AAL have a low cost as only the Therefore,a low-cost,low-power technology is desperately RFID readers are required to be deployed.RTL has a median required for activity sensing.Technological advances in RFID deployment cost since the reference tags are necessarily to has provided the opportunities for activity sensing in an eco- be deployed in advance.FL has a high deployment cost as nomically attractive approach.By exploiting the phenomenon collecting the training data of RSSI fingerprints is rather time- that RSSI changes significantly when an object is passing consuming and inconvenient.In regard to the hardware cost, by,Liu et al.propose to use RFID tag arrays for activity RL,FL and RTL have a low cost in hardware as conventional sensing in a device-free approach,i.e.,the tracking objects off-the-shelf RFID readers can effectively collect the RSSI do not need to attach any transmitters or receivers,such as for localization.PL and AAL introduce a high cost since they tags or readers [84].By developing a practical fault-tolerant require precise instrument to obtain the measurements like the method,they offset the noise of RF tag data and mine frequent phase and angle of signals. trajectory patterns as models of regular activities.Zhang et al. further propose TASA,a tag-free activity sensing using RFID tag arrays for location sensing and route tracking [85].Being different from the previous scheme [84].TASA uses passive F.Summing Up Challenges and Opportunities tag arrays together with a few active reference tags,instead of all active tags.Moreover,by reducing noise in the readings According to the above recent research progress,we sum- of passive RFID tags,TASA is much effective for locating marize the challenges and opportunities for RFID localization multiple moving objects. and activity sensing in Table VIII.We believe that,by suffi- ciently exploring the underlying opportunities in localization E.Analysis and activity sensing,the challenges can be gradually over- In order to get a better understanding of the features for come to finally achieve accurate and real-time perception of the aforementioned localization schemes,we compare their locations and activities.XIE et al.: MANAGING RFID DATA: CHALLENGES, OPPORTUNITIES AND SOLUTIONS 13 TABLE VII COMPARATIVE STUDY OF THE LOCALIZATION SOLUTIONS Accuracy Time-efficiency Deployment Cost Hardware Cost Range-based localization using RSSI (RL) Low High Low Low RSSI fingerprint-based lo￾calization (FL) High High High Low Reference tag-based local￾ization using RSSI (RTL) Median Low Median Low Phase-based localization (PL) High Median Low High Angle of Arrival based lo￾calization (AAL) High Median Low High RFID faults greatly affect the localization accuracy, in order to tackle this problem, Zhu et al. propose an effective fault￾tolerant RFID reader localization approach suitable for the above situations, and illustrate how to measure the quality of a localization result [82]. In order to deal with the noisy RFID readings gathered in object tracking applications, Yang et al. propose a hybrid method for tracking mobile objects with high accuracy and low computational cost [83], such that the computational cost of filtering the noises can be greatly reduced. D. Activity Sensing Activity sensing has drawn many attentions in recent years, and it has yielded lots of research results. Most of them are related to vision-based schemes and surveillance technologies. The surveillance technologies are further classified into the Bluetooth, infrared, ultrasonic and sensor based solutions. These solutions mainly require fairly expensive equipments to capture scenarios as video or other sensor data, moreover, a lot of energy is expended for these equipments during the procedure of activity sensing. These issues greatly limit the range of applications, as most pervasive applications cannot afford such a huge overhead in terms of both deployment cost and energy consumption. Therefore, a low-cost, low-power technology is desperately required for activity sensing. Technological advances in RFID has provided the opportunities for activity sensing in an eco￾nomically attractive approach. By exploiting the phenomenon that RSSI changes significantly when an object is passing by, Liu et al. propose to use RFID tag arrays for activity sensing in a device-free approach, i.e., the tracking objects do not need to attach any transmitters or receivers, such as tags or readers [84]. By developing a practical fault-tolerant method, they offset the noise of RF tag data and mine frequent trajectory patterns as models of regular activities. Zhang et al. further propose TASA, a tag-free activity sensing using RFID tag arrays for location sensing and route tracking [85]. Being different from the previous scheme [84], TASA uses passive tag arrays together with a few active reference tags, instead of all active tags. Moreover, by reducing noise in the readings of passive RFID tags, TASA is much effective for locating multiple moving objects. E. Analysis In order to get a better understanding of the features for the aforementioned localization schemes, we compare their performance in terms of accuracy, time-efficiency, deploy￾ment cost and hardware cost. Here, the accuracy denotes the precision of localization, the time-efficiency denotes the locating speed, the deployment cost denotes the human cost of deploying the localization system and getting the training data, the hardware cost denotes the fixed cost of the RFID readers and tags in the localization systems. Table VII depicts the performance comparison of the lo￾calization solutions. In order to make it clear and concise, in the following we respectively use the abbreviations RL, FL, RTL, PL and AAL to denote the corresponding location schemes. In regard to the accuracy, FL, PL and AAL all achieve high accuracy in localization, the localization errors can conventionally be limited to at most 0.5m. RTL achieves a median accuracy by limiting the errors to at most 1m. RL achieves a low accuracy with the localization errors of at most 5m. In regard to the time-efficiency, both RL and FL achieve high-efficiency since they only require to obtain the RSSI from the target tag. PL and AAL achieve median-efficiency as more refined data are required and more computations are involved in the localization. RTL achieves low-efficiency in time, because the RSSIs from multiple tags are required to be obtained, which can be very time consuming. In regard to the deployment cost, RL, PL and AAL have a low cost as only the RFID readers are required to be deployed. RTL has a median deployment cost since the reference tags are necessarily to be deployed in advance. FL has a high deployment cost as collecting the training data of RSSI fingerprints is rather time￾consuming and inconvenient. In regard to the hardware cost, RL, FL and RTL have a low cost in hardware as conventional off-the-shelf RFID readers can effectively collect the RSSI for localization. PL and AAL introduce a high cost since they require precise instrument to obtain the measurements like the phase and angle of signals. F. Summing Up Challenges and Opportunities According to the above recent research progress, we sum￾marize the challenges and opportunities for RFID localization and activity sensing in Table VIII. We believe that, by suffi- ciently exploring the underlying opportunities in localization and activity sensing, the challenges can be gradually over￾come to finally achieve accurate and real-time perception of locations and activities. This article has been accepted for inclusion in a future issue of this journal. Content is final as presented, with the exception of pagination