3D Indoor localization based on spectral clustering and weighted backpropagation neural networks

With the rapidly growing interest for indoor location-based services (LBSs), a number of localization techniques have been developed especially for 3D environment. While high accuracy could be achieved by most of available techniques, expensive devices of superb processing ability are basically required. On this account, a backpropagation neural network (BPNN) based method is applied for 3D indoor localization in this work, where only received signal strength (RSS) is required for the fingerprint database construction without additional hardware. Nevertheless, without considering the correlation of wireless channels, the BPNN based method cannot achieve desirable localization accuracy for 3D environment. For this reason, we propose a spectral clustering and weighted backpropagation neural network (SWBN) method to enhance the localization accuracy. Specifically, considering the spatial channel correlation, RSS is first partitioned into several groups using a tailored spectral clustering based method. Afterward, the weighted BPNN algorithm is adopted for location estimation. Experimental results show that SWBN could reduce localization median error by 36.21% and 11.38%, respectively, compared to the state-of-the-art k-nearest-neighbor method and BPNN method. Besides, the training time is significantly reduced by 41.48% compared to BPNN.