A Fingerprinting Indoor Localization Algorithm Based Deep Learning

Fingerprinting in essence uses a machine to infer physicals locations from radio map data. This machines are usually either probabilistic and neural networks consisting of one layer. In this propose we use deeper machines (DNN, DBN and GB-DBN) to increase the estimation accuracy and reduce generalization error on dynamic indoor environment. Also we investigated the impact of pre-training algorithm on fingerprinting indoor location systems. Experimental results demonstrate that deep models provide an efficient generalization performance on indoor environments. They have the disadvantage that demand high processing resources when they are trained on off-line phase, however, deep models are swift to predict during on-line phase.