A WiFi fingerprint based high-adaptability indoor localization via machine learning

For received signal strength (RSS) fingerprint based indoor localization approaches, the localization accuracy is significantly influenced by the RSS variance, device heterogeneity and environment complexity. In this work, we present a high-adaptability indoor localization (HAIL) approach, which leverages the advantages of both relative RSS values and absolute RSS values to achieve robustness and accuracy. Particularly, a backpropagation neural network (BPNN) is devised in HAIL to measure the fingerprints similarities based on absolute RSS values. With this aid, the characteristics of the applied area could be specially learned such that HAIL could be adaptive to different environments. The experiments demonstrate that HAIL achieves high localization accuracy with the average localization error of 0.87m in the typical environments. Moreover, HAIL has the minimum amount of large errors and decreases the average localization error by about 30%∼50% compared with the existing approaches.