RFID-based Section Shape Recognition

Currently RFID based spatial feature sensing technology has been developed rapidly.For example,the location accuracy of RFID based indoor localization system is of the order of centimeters and the mean three-dimensional orientation estimation accuracy is also less than4°.There are a number of RFID based spatial feature sensing systems deployed as applications like baggage sorting system in express station and airport,which unconsciously influence the daily life of people.However,these systems are only capable of sensing and applying simple spatial features such as location and orientation.Before they can provide more complex spatial feature like section shape,many challenging problems must be solved.The majority of existing radio-frequency signal based target imaging systems only find the general location of target and detect its blurred outline.One of the latest studies recognizes more accurately the shape of target object making use of the penetrability of RFID signal,but it is sensitive to the internal material and structure of target and unable to reconstruct complex shape like a concave polygon.Accurate section shape recognition of target object can be now only achieved by expensive dedicated devices.As a much cheaper auxiliary means,RFID based SSR technology,whose foundation is the reflectivity of RFID signal,is proposed for accurate shape recognition of horizontal section of target object.The main contributions of this paper are listed as follows.(1)Firstly ISR,a physical quantity robust to multipath effect and measured deviation caused by hardware,is able to be calculated from raw measured data after the RFID device arrangement and the measurement method are carefully designed,so that the shape information of horizontal section of target object can be extracted from the scene full of wireless channels.(2)Next the one-dimensional peak finding based fast AoA estimation algorithm is used for stable and efficient AoA estimation.The design of this algorithm is motivated by a theoretical result that the SSR RFID system has a more accurate AoA estimation than distance estimation.Another result is the signal interference problem,which degrades the AoA estimation accuracy and arises when the number of reflected signals from target object is more than one.(3)Then it is necessary to move target object or RFID system to satisfy the spatial diversity of measured value,which makes it possible to improve distance estimation.Moving brings AoA dataset of time.The AoA-time dataset is separated into one or more sequential AoA clusters from different reflectors via spectral clustering.Data smoothing and data generation are also applied to dealing with the signal interference problem.(4)Finally an optimization model is built for aggregation of sequential AoA data.On the basis of the geometrical relationship between RFID system and target object,an equation is relaxed as objective function and the others are transformed into constraints.Each size of horizontal section of target object is directly reconstructed by the optimal solution of this model.The experiments are conducted in a common meeting room for the evaluation of key technologies and performance of SSR prototype.The first experiment is about ISR verifying its robustness to multipath effect and measured deviation caused by hardware.The second one checks the comparison of AoA estimation accuracy of SSR to its distance estimation accuracy and analyzes the correlation factors in detail.Then evaluate the performance of one-dimensional peak finding based fast AoA estimation algorithm and MUSIC algorithm in the cases of only one reflected signal from target and two signals.The last one shows the effects of spectral clustering,data smoothing,data generation and the optimization model with respect to the division of AoAtime dataset,the elimination of signal interference and size reconstruction.The result indicates that the mean AoA estimation error is about 1.5° without signal interference.In terms of size reconstruction,rotation angle estimation error is less than 4° and the mean minimal offset distance is at best not exceeding 10 cm.