Efficient and Scalable Calibration-Free Indoor Positioning Using Crowdsourced Data

Abstract

With the development of Internet of Things (IoT), a variety of crowdsourced data becomes available. In this article, we propose to utilize the crowdsourced WiFi received signal strength (RSS) data to perform indoor positioning. In many instances, the size of crowdsourced WiFi RSS data is potentially large, and the ground truths of the corresponding RSS fingerprints are unavailable. Therefore, it is challenging to construct the associated radio map. In the proposed method, a heuristic geometrical algorithm is developed to convert the RSS data into pairwise distances among the fingerprints. Based on these pairwise distances, multidimensional scaling (MDS) is then applied to compute the positions of all the fingerprints, thereby building a radio map. We show that the accuracy of the proposed method is only reasonably lower than a state-of-the-art calibration-based method. Further, we present a ${k}$ -means clustering-based region partitioning method that partitions the large crowdsourced data set effectively. The parallel processing on the partitioned data and computational simplicity of MDS result in significantly shorter runtime for the proposed method than the previous optimization-based methods.