Reduce Redundancies: Signal-based Clustering of Large-scale Fingerprint Data

For Bluetooth- or WIFI-based localization, fingerprinting data plays an important role. Mobile devices compare their received signals with recorded signals on a map (so-called reference points) and derive their most like location from that. Obviously, this method requires an elaborate offline phase to record the reference points. If this data set is too small and not dense enough, the localization accuracy is low, too. However, by just increasing the number of recorded data points, storage and comparison costs on the mobile device are also increased. Hence, the goal of this work is to find the best reference points within large-scale sets of fingerprinting data based on clustering. We present different novel algorithms for signal-based clustering and compare them with existing work. An extensive evaluation on real-world data sets shows that our approach can reduce the data set size up to 90% while keeping a mean accuracy of 1.0m in the experiments in the real world.