A Noise Reduction Fingerprint Feature for Indoor Localization

Indoor localization has received extensive attention in recent years. Due to the complexity of the indoor environment, fingerprint positioning methods have better performance than the methods based on propagation models. In the WiFi network, the received signal strength (RSS) is usually used as the fingerprint feature. Due to the influence of noise in the environment, there is the error between the fingerprint information and the online data, which results in deviation of the localization results. We proposed a novel noise reduction fingerprint feature which calculates the difference of RSS at the same location and combines it with the original RSS and then obtains the fingerprint feature through the principal component analysis, which can improve the robustness of the Wi-Fi signal and reduce the influence of the environmental noise. In order to valid the performance of the proposed fingerprint feature, we introduced the recurrent neural network based indoor localization algorithm. The algorithm takes the correlation of historical locations into consideration and establish the relationship between the characteristics of the wireless signal and location by recurrent neural network. The experimental result shows that the proposed fingerprint feature has better performance than conventional fingerprint feature.