novel localization method for moving object using integration experiments.In our experiments,we use the Alien-9900 reader of passive RFID tags and scene analysis technique [15]. and Alien-9611 linear antenna with a directional gain of 6dB. LANDMARC [16]is a tag localization prototype in indoor The 3dB beamwidth is 65 degrees.The RFID tags used environment.By utilizing extra reference with fixed reference are Alien 9640 general-purpose tags which support the EPC tags to help location calibration,it can increase location accu- CIG2standards.We place the tags on a shelf with 5 rows,the racy without deploying large numbers of RFID readers.Zhu et distance between two rows is 60cm. al.propose a fault-tolerant RFID reader localization approach In the following experiments,we vary the tag density from to solve the problem of frequent occurred RFID faults [17].2 to 10 tags per meter,while adjusting the reader's power from Xiao et al.propose a novel environmental-adaptive indoor 15.7dBm to 30.7dBm.Unless otherwise specified,in default positioning approach using RSSI [18].The signal propagation situation we fix the reader towards the center of the shelf, model and model parameters are updated in a closed-loop scan the tags for 50 query cycles repetitively.And the distance feedback correction manner. between the antenna and the shelf is 1.5m. III.PROBLEM FORMULATION Shopping malls,warehouses and public libraries,may have ◆10ta3/m ◆5tag/m massive RFID tags deployed in the environment(There are 3tags/m many objects or books attached with tags on the shelves).Our -1.5tags/m objective is to use these existing tags to localize the mobile user equipped with an RFID reader,e.g.,the reader may be attached to a shopping cart.When the user moves with the reader in these environments,the reader can interrogate the 面司 tags within the scanning range and get their identities for localization. Fig.3.Detection Regions'radius Fig.4.Relationships among Tag size,Power and Tag density 600 00网 Fig.1.Localization process Fig.2.Tags on the shelf Fig.5.Scanning TIme Fig.6.RSSI distribution(28.7dBm) As shown in Fig.1,there are many shelves in the localization environment.The star is the target needed to be located.As TABLE I the shelves'locations are fixed,we can use one-dimension TABLES OF TRAINING DATA SETS coordinate system to describe the target's position.In the Table Name Description figure,the yo is known.By estimating the position z in the Ta Detection regions'radius in different tag density and power. aisle,we finally get the exactly location of the target. Relationships among tags size. As shown in Fig.2,the shelf is divided into several blocks T Power and Tag density. and each block contains multiple objects represented by small Te Relationship between detected tags size and scanning time. cubes.The blocks are distinguished by their colors in the ta Relationship between RSSI value and distance figure.The objects in one block belong to the same category, to the center of the detection region. which is represented in tag ID.The width of the blocks can also be different from each other.We describe a block as 1)Tag density effects:In the same scanning range,the [i,s,xi.e,li].i.s,xi,e are the start point and end point of density p of the tags is the main influencing parameter.A the block,l;represents the layer of the shelf,and the height high density p causes more tags to be identified in the range. of each shelf layer is h.The exact position of each object Meanwhile,as the tag density increases,the major detection in the block is unknown,while the location of each block is regions radius gradually decreases.Fig.3 shows the relation- known. ship of detection region,reader's power,and tag density.We can find that when the power is fixed,as the tag density IV.OBSERVATIONS FROM THE REALISTIC EXPERIMENTS increases.the width of the detection region decrease.The Because of the issues like path loss,energy absorption Fig.4 shows the relationship of the number of identified tags, and mutual interference,the RSSI distribution are always not the reader's power,and the tag density.Based on these two idealized.In order to get these information in the realistic relationships,we make two training data sets Ta and T by environment,we provide several observations from realistic experiments.Then we can compute the tag density and thenovel localization method for moving object using integration of passive RFID tags and scene analysis technique [15]. LANDMARC [16] is a tag localization prototype in indoor environment. By utilizing extra reference with fixed reference tags to help location calibration, it can increase location accu￾racy without deploying large numbers of RFID readers. Zhu et al. propose a fault-tolerant RFID reader localization approach to solve the problem of frequent occurred RFID faults [17]. Xiao et al. propose a novel environmental-adaptive indoor positioning approach using RSSI [18]. The signal propagation model and model parameters are updated in a closed-loop feedback correction manner. III. PROBLEM FORMULATION Shopping malls, warehouses and public libraries, may have massive RFID tags deployed in the environment(There are many objects or books attached with tags on the shelves). Our objective is to use these existing tags to localize the mobile user equipped with an RFID reader, e.g., the reader may be attached to a shopping cart. When the user moves with the reader in these environments, the reader can interrogate the tags within the scanning range and get their identities for localization. Fig. 1. Localization process Fig. 2. Tags on the shelf As shown in Fig.1, there are many shelves in the localization environment. The star is the target needed to be located. As the shelves’ locations are fixed, we can use one-dimension coordinate system to describe the target’s position. In the figure, the y0 is known. By estimating the position x in the aisle, we finally get the exactly location of the target. As shown in Fig. 2, the shelf is divided into several blocks and each block contains multiple objects represented by small cubes. The blocks are distinguished by their colors in the figure. The objects in one block belong to the same category, which is represented in tag ID. The width of the blocks can also be different from each other. We describe a block as [xi,s, xi,e, li ]. xi,s, xi,e are the start point and end point of the block, li represents the layer of the shelf, and the height of each shelf layer is h. The exact position of each object in the block is unknown, while the location of each block is known. IV. OBSERVATIONS FROM THE REALISTIC EXPERIMENTS Because of the issues like path loss, energy absorption and mutual interference, the RSSI distribution are always not idealized. In order to get these information in the realistic environment, we provide several observations from realistic experiments. In our experiments, we use the Alien-9900 reader and Alien-9611 linear antenna with a directional gain of 6dB. The 3dB beamwidth is 65 degrees. The RFID tags used are Alien 9640 general-purpose tags which support the EPC C1G2standards. We place the tags on a shelf with 5 rows, the distance between two rows is 60cm. In the following experiments, we vary the tag density from 2 to 10 tags per meter, while adjusting the reader’s power from 15.7dBm to 30.7dBm. Unless otherwise specified, in default situation we fix the reader towards the center of the shelf, scan the tags for 50 query cycles repetitively. And the distance between the antenna and the shelf is 1.5m. Fig. 3. Detection Regions’ radius Fig. 4. Relationships among Tag size, Power and Tag density 0 50 100 150 0 0.5 1 1.5 2 2.5 The number of identified tags Scanning Time (s) Fig. 5. Scanning TIme Fig. 6. RSSI distribution(28.7dBm) TABLE I TABLES OF TRAINING DATA SETS Table Name Description Ta Detection regions’ radius in different tag density and power. Tb Relationships among tags size, Power and Tag density. Tc Relationship between detected tags size and scanning time. Td Relationship between RSSI value and distance to the center of the detection region. 1) Tag density effects: In the same scanning range, the density ρ of the tags is the main influencing parameter. A high density ρ causes more tags to be identified in the range. Meanwhile, as the tag density increases, the major detection regions radius gradually decreases. Fig.3 shows the relation￾ship of detection region, reader’s power, and tag density. We can find that when the power is fixed, as the tag density increases, the width of the detection region decrease. The Fig.4 shows the relationship of the number of identified tags, the reader’s power, and the tag density. Based on these two relationships, we make two training data sets Ta and Tb by experiments. Then we can compute the tag density and the