A Practical Floor Localization Algorithm Based on Multifeature Motion Mode Recognition Utilizing FM Radio Signals and Inertial Sensors

Floor localization plays a crucial role in multi-storey indoor positioning systems. Wi-Fi enabled system and barometer-based system are two well-known systems in floor localization. However, a practical floor localization system should consider the absence of infrastructure, the pretrained database and the influence of complex environment. Inertial sensors-based systems realize floor localization by classifying pedestrian's motion states, which can avoid the above problems. Nevertheless, the accuracy is limited because of the stochastic noise introduced by device measurement. Besides, only single acceleration feature is utilized in most of the existing systems. Fusion of multiple inertial sensors and other signals such as frequency-modulated (FM) radio signals is a feasible solution considering both practicality and accuracy. In this paper, a multi-feature floor localization algorithm using MEMS inertial sensors and a FM radio receiver is designed. Firstly, based on the MEMS inertial sensors and FM radio signals, we select features which are not affected by types or sites of stairs for the classification of plane walking, going upstairs, and going downstairs moments. Secondly, random forest classifiers are introduced to obtain accurate classification results by combining the above features. Floor is located on the basis of classification results. Field experiments have been conducted at four sites in two office buildings to evaluate the performance of the proposed algorithm. The results demonstrate that the multi-feature approach achieves 96.9% floor localization accuracy, which is 38.4% higher than the method using only single acceleration feature and 24.3% higher than the multi-feature algorithm without FM features.