Off-the-shelf indoor localization system using radio frequency for wireless local area network

The indoor localization system becomes a substantial issue in recent research, especially in terms of the accuracy. Location based services have been used in many mobile applications as well as wireless sensor networks. High accuracy and fast convergence are very important issues for a good localization system. However, the type of obtained received signal strength (RSS) data is very important in order to get high accuracy. Recently, several of indoor localization techniques that are based on signals of wireless local area network (WLAN) become a substantial issue in recent research. In this research, a fingerprinting-based location algorithm is applied in indoor environments using WLAN. The location fingerprinting algorithm consists of two phases: offline phase and online phase. In the offline phase the reference points (RPs) are collected at certain places in the experimental testbed. The measurement campaign is conducted by using developed Wi-Fi scanner software. During the offline phase an extensive study is performed on the RSS properties for indoor environment such as duration effects, RSS stationary, RSS dependency and a user’s presence. In the online phase, the proposed model infers the unknown locations based on the RPs available in the radio map. The user location is inferred based on three dimensional (3-D) Bayesian graphical model using the OpenBUGS program. The inference of user location in the environment is investigated and compared to the actual location. Besides, the numbers of iterations are examined in order to show its effectiveness on the proposed model. It shows that the model is converged at a level of 100000 iterations. Thus, the best choice of number of iterations for the proposed model is 100000 since there is no improvement if the number of iterations increases. Finally, the proposed Bayesian graphical model based on fingerprinting location algorithm is compared with Madigan model. The proposed Bayesian graphical model and Madigan model achieved an average accuracy of 2.9 and 3.8 meters for 50 RPs, respectively. Besides, the proposed model is off-the-shelf which does not require any additional hardware to integrate to the proposed model. The proposed system is enhanced further by using offline clustering (OC) algorithm to reduce the data size of radio map and improve the system’s accuracy. In the first stage, the OC tries to reduce the number RPs in the radio map by grouping sets of RPs that are close to each other into one cluster. In the second stage, one or more cluster joins together based on the distance of signal space between adjacent clusters. The proposed OC algorithm slightly reduced the localization error to 2.4 meters, while it significantly reduced the data size of radio map by 68%