In this paper we present extensive experimental results of location fingerprinting with passive UHF radio-frequency identification (RFID). As recent passive RFID hardware provides information about received signal strength (RSS), we evaluate its usefulness in the context of fingerprinting based on classical vector similarity measures. We analyze the impact of decisive parameters of the applied approach and also select them automatically via cross-validation, including the most appropriate similarity measure. A further novelty is an RSS thresholding mechanism which reduces the computational demands of comparing fingerprints. This technique is especially useful in surroundings which are densely equipped with RFID tags, such as future supermarkets or logistic centers. We conducted realworld experiments with a mobile robot and two different RFID readers. Results are reported both for global localization in each time frame and for time-filtered position tracking. We provide the experimental data of this work for download. Details on the experimental setup as well as downloadable datasets can be obtained at http://www.cogsys.cs.uni-tuebingen.de/datasets/rfid-ta2011/.
@inproceedings{vorst2011rfidta, author = {Philipp Vorst and Artur Koch and Andreas Zell}, title = {Efficient Self-Adjusting, Similarity-based Location Fingerprinting with Passive {UHF} {RFID} }, booktitle = {IEEE International Conference on RFID-Technology and Applications (RFID-TA2011)}, year = {2011}, pages = {160--167}, address = {Sitges, Barcelona, Spain}, month = {September 15-16}, publisher = {IEEE}, abstract = {In this paper we present extensive experimental results of location fingerprinting with passive UHF radio-frequency identification (RFID). As recent passive RFID hardware provides information about received signal strength (RSS), we evaluate its usefulness in the context of fingerprinting based on classical vector similarity measures. We analyze the impact of decisive parameters of the applied approach and also select them automatically via cross-validation, including the most appropriate similarity measure. A further novelty is an RSS thresholding mechanism which reduces the computational demands of comparing fingerprints. This technique is especially useful in surroundings which are densely equipped with RFID tags, such as future supermarkets or logistic centers. We conducted realworld experiments with a mobile robot and two different RFID readers. Results are reported both for global localization in each time frame and for time-filtered position tracking. We provide the experimental data of this work for download. Details on the experimental setup as well as downloadable datasets can be obtained at \url{http://www.cogsys.cs.uni-tuebingen.de/datasets/rfid-ta2011/}.}, doi = {10.1109/RFID-TA.2011.6068632}, ee = {http://dx.doi.org/10.1109/RFID-TA.2011.6068632}, isbn = {978-1-4577-0028-6}, url = {http://www.cogsys.cs.uni-tuebingen.de/publikationen/2011/vorstkoch2011rfidta.pdf} }