A Hidden Markov model for indoor tracking Based on Bluetooth fingerprinting and Grid filtering

With the popularization of smartphone and the development of mobile internet, people’s demand on location-based services has increased. Indoors, users are often repeating the same movements, such as walking between main door, office, rest rooms. As a result, massive of pedestrian tracks have been produced. Fingerprint location algorithm based on Wi-Fi or Bluetooth is a popular method in indoor locating. However, this method has many weaknesses including unsteadiness and low robustness, because it generally uses the mean value and its variance of single point Rssi to locate. Grid filters the incorrect positions returned from indoor. Localization, especially under signal fluctuation or fingerprint ambiguity. The Hidden Markov model and Grid filtering of this approach would achieve a higher position accuracy and efficiency. Map information is often available and may contribute to location filtering. Recently, a number of researchers take the position sequence information into consideration. Xiaoguang(2014) had found a fine-grained walk pattern of indoor pedestrians the reliability of the reports. Zhou(2014) had proposed activity sequence based indoor pedestrian localization using smartphones. Zhang(2015) proposed a wireless positioning method based on Deep Learning arm to deal with the variant and unpredictable wireless signals. Jimmy(2013) proposed directional HMM algorithm which can learn user habits and improve the accuracy of indoor localization system. However, the HMM models are trained with the trajectory and the HMM indoor position algorithm relays on the correct and certain data(He, S., & Chan 2015). Borriello(2003) proposed Bayesian Filtering which include the grid filtering to estimate the position. But the HMM model still cannot deal well with the ambiguities resulting and bring the amount of calculation. Therefore, a key challenge here is how to deal with the ambiguities resulting from fingerprinting and the data fusion of big spatial data and limited topology information. LBS 2016