Smartphone-Based Indoor Localization Using Machine Learning and Multisource Information Fusion
Jun Yan, Zheng Huang, Xiaohuan Wu
Abstract
A two-phase smartphone localization technique that uses received signal strength indicator (RSSI) fingerprints of long term evolution (LTE) signal, Bluetooth signal, Wi-Fi signal and the internal camera sensor is proposed for indoor environments. It contains: (1) coarse localization: region determination by LTE and Bluetooth signal and (2) refined localization: position estimation by camera image and Wi-Fi signal. To maximize efficiency we develop data fusion algorithm, aiming to (a) combine the RSSI measurement of Bluetooth and LTE signal to form coarse localization fingerprint. (b) transform the RSSI measurements of Wi-Fi signal into image representation by linear mapping method. (c)fuse the camera image and Wi-Fi radio image by the pixel level image fusion and pyramid decomposition method. The proposed solution is unique in that its offline phase exploits support vector machine for all regions to generate region classification functions. And for each region, it exploits convolution neural network to generate position regression function. The online phase executes a coarse localization step to estimate the region by using the region classification functions and a refined step to estimate the position by using the position regression function. Experiment results show that the proposed algorithm outperforms existing schemes.