Indoor Motion State Recognition using Denoising Autoencoders on Smartphone

The interest of indoor movement state recognition has been recently extensively evaluated by several research teams. This paper fully explores the signals from all the sensors in a smartphone and develops a novel deep neural network-based model. In the off-line training phase, this paper firstly proposes a matrix to pre-processing the raw data from a smartphone to overcome its redundancy and complexity. In a second place, this paper managed to go through a stacked denoising autoencoder neural network, resulting in an output layer with seven different indoor movement states. The trained network is then implemented on a smartphone and evaluated with online tests. Extensive experimental results demonstrate that the proposed method achieves 20% higher identification accuracy than other existing methods with respect to various indoor motion status (e.g., walking up and down stairs, taking elevator up and down).