Unsupervised Learning of Radio Map from Imbalanced Crowd-sourced Trajectories

The indoor localization based on WiFi fingerprints has attracted great attention, where the most important but challenging work is the construction of the radio map without the need for floor maps. The rapid popularization of mobile and wearable devices makes the collection of massive trajectory data by crowd-sourcing convenient and cost-effective. However, the crowd-sourced trajectory data are with low quality, which usually have an imbalanced distribution due to the presence of hot spots and restricted zones in the building and the random movements of participants. To address this challenge, in this paper, we propose the Unsupervised Learning Radio Map from Imbalanced Crowd-sourced Trajectories scheme (MapICT) to build radio map. First, the Hierarchical Density-Based Spatial Clustering of Applications with Noise (HDBSCAN) algorithm is suggested to solve the problem of imbalanced distribution of trajectory data. Second, smoothing algorithm is proposed to optimize the clustering result. The MapICT scheme is proven feasible and effective through the simulation experiments and experiments in actual environment without the floor maps, where the locating accuracy of radio map constructed is 9.86% higher than existing methods.