LAUSR

Fingerprint-based indoor localization methods have become an important technology because of their wide availability, low hardware costs, and the rapidly growing demand for location-based services. However, it is low precision of positioning and time-consuming for retraining the model when the fingerprint database has changed with new input samples. In this paper, we propose a novel intelligence localization scheme utilizing incremental learning without retraining models based on channel state information (CSI), namely ILCL. CSI phase data are extracted through a modified device driver, and we convert them into CSI images, which are the input to a convolutional neural network for training the weights in the offline stage. The estimated location is obtained by a probabilistic method based on a broad learning system (BLS) that can continue to train rapidly on new input data in the online stage. The ILCL architecture can be characterized as “deep” and “broad” and can further extract features. Experimental results confirm the superiority of ILCL compared with five existing algorithms in two real-world indoor environments with a total area is over 200 ${m}^{2}$.