A Novel Tensor Completion Based Indoor Positioning Fingerprint Recovery Method in Mobile Crowdsensing Networks

As an emerging paradigm, Mobile CrowdSensing (MCS) network based fingerprint positioning technology can implement the site survey cheap and fast, and provide a more reliable location-based service for users. However, limited by the differences among individuals, the fingerprints collected from MCS network are troubled by fingerprint ineffectiveness or element missing problems. In order to ensure the availability of fingerprints, in this paper we propose a novel Low-rank and Sparse representation based Tensor Completion (LSTC) fingerprint recovery method to recover the ineffective fingerprints and the missing fingerprint elements accurately in MCS network. Specifically, the tensor low-rank characterization is used to exploit the global structure of the location fingerprints, and the tensor sparsity characterization is used to exploit the local structure of the location fingerprint elements. To promote the global optimal solution solving rapidly, a weight compensation scheme is proposed to fill the convex relaxation gap caused by the approximation of low-rank and sparsity. Meanwhile, to represent the local sparse characterization effectively, orthogonal dictionary learning is introduced and integrated into LSTC. The experimental results show that by using the tensor global and local structure information properly, the proposed LSTC can recover the fingerprint effectively, without jeopardizing the positioning accuracy.