FILM: Fine-Grained Book Localization with Mobile RFID Scanning.
Xingyu Bian,Xia Wang,Wei Cheng,Xingyu Chen,Jia Liu,Lijun Chen
DOI: https://doi.org/10.1109/icpads47876.2019.00088
2019-01-01
Abstract:Misshelving is an age-old problem in a library. If this happens, users can hardly find their wanted books, leading to a waste of book resources. In this paper, we investigate the problem of book localization through mobile RFID scanning. According to extensive experiments, we find that the phase data (that are used to do localization) beyond the reader antenna's main lobe are noisy; taking these data as the input of book localization is likely to lower the localization accuracy. However, it is not easy to figure out which data are noisy. To address this problem, we propose a fine-grained localization method called FILM, which can achieve cm-level book localization through the following three steps. First, it collects the phase profile for each tagged book as is. Second, it sophisticatedly chooses three samples from the phase profile and locates the tagged book by modeling the three samples. Since there are a great number of 3-combinations, we can get many candidates of the book localization. Third, it uses the Gaussian kernel density model to determine whether the candidates are good or not. The basic idea is that, the noisy data give rise to uncertain estimates, whereas the clean data will always get an estimate close to the ground truth. Therefore, the position with the highest density is the most likely to be the tag's position. By this means, the noisy data do not contribute to the final estimate, which improves the localization accuracy. We setup a prototype of FILM and experiment results show that FILM is superior to the state-of-the-art, with a small localization error of 1.3cm.