Research On Sparse Fingerprint-based Indoor Localization

Indoor localization is of great importance for a range of mobile applications, thus motivates researchers make continual efforts in the past decades. Although GPS(Global Positioning System)can obtain location of room-level accuracy outdoors, it performs poorly under indoor environment since lacking line of sight with orbiting satellites.RSSI(Received Signal Strength Indicator)-based localization take full advantage of the existing infrastructures thus avoid the expenditure of speci?c deployment. However, existing RSSI-based indoor localization scheme either relies on the basic assumption that APs are pervasive in the environment or the priori knowledge of every AP’s location which is not always possible in the real world. But the typical scenario like airport or railway stations where APs are fractionally covered either due to the limited deployment or biased deployment which result in sparse ?ngerprint. As the experiment we have conducted indicated, the performance behaviours poorly when few APs fed to the RSSI-based ?ngerprint localization scheme.PDR(Pedestrian Dead Reckoning) is a potential technique to calculate the position under ?ngerprint sparse environment since it only use the inertial sensors embedded in smart-phones. In this paper, we introduce a novel, effective and accurate way to detect steps based on the off-the-shelf smart-phones’ quality limited commercial inertial sensor readings thus providing a Dead Reckoning sub-system. Together with fusing the occasionally sensed RSSI from ambient APs with the sparse ?ngerprint by applying particle ?lter, we are able to alleviate the effect of the accumulate error and improve the localization accuracy.We implement the designed system and conduct extensive experiments on Android-based phones both in ?ne-grain and coarse-grain testbed. The step detection algorithm exceeding its opponents in both performance and e?ciency and the experimental result shows nearly 100% accuracy. Evaluation results of localization accuracy indicate that our localization scheme can achieve high accuracy even in RSSI ?ngerprints sparse environment, and performs well in both ?ne-grain and coarse-grain testbed.