Reducing fingerprint collection for indoor localization

A typical WiFi-based indoor localization technique estimates a device’s location by comparing received signal strength indicator (RSSI) against stored fingerprints and finding the closest matches. However, the collection of fingerprints is notoriously laborious and costly. It is challenging to reduce fingerprint collection and recover missing data without introducing significant errors. In this article, a novel approach based on compressive sensing is presented for recovering absent fingerprints. The hidden structure and redundancy characteristics of fingerprints are revealed in a merging matrix. The spatial and temporal correlations of fingerprints result in a small rank of the merging matrix. The Sparsity Rank Singular Value Decomposition (SRSVD) method is used to effectively reduce the interference caused by the multipath effect of the WiFi signal. We further propose to combine SRSVD with the K-Nearest Neighbor (KNN) algorithm to deal with missing columns or rows in the matrix. Experimental results show that with only half of the fingerprints, our approach can recover all the fingerprint information with error rate below 6.6%. Even with only 5% of the data, the approach can recover the information with error rate below 14%, without loss of localization accuracy.