| With the development of wireless communications technology and pervasive computing, location-based services(LBS) have been applied extensively in daily life, and location technology is an essential prerequisite and core link to realize LBS. But the wireless signals required by location are easily influenced by attenuation in the propagation process and complex wireless propagation environment, which makes outdoor localization technology hardly be applied to indoor applications. Therefore, indoor localization becomes a hot issue and many systems have appeared in the course of study. At present, location fingerprinting based indoor localization has raised widespread concerns in both academic and industrial fields, due to its advantages such as low costs, convenient deployment, no need of access point location and extra hardware support. However, the complexity and particularity of indoor environment makes the positioning accuracy be not highly enough, and the huge fingerprinting database leads to high computational complexity. In this paper, we take use of methods of Voronoi diagram and naive Bayesian method to improve accuracy, and also propose a new fingerprinting database construction method to reduce the computational complexity.Firstly, we propose a Voronoi analytical model based on graph theory, and apply this model to analyzing the fingerprint structure, yielding proximity information and computing the centroid of the Voronoi vertex in the Voronoi region. Furthermore, we compare the measured location with the actual location. Based on the comparison results, we select the smallest Euclidean distance between the two locations as the approximation of the actual location. In order to validate the performance of analytical model on efficiency and reliability, we conduct an extensive experiment in an indoor parking lot, where it is convenient to deploy the access points(APs). The simulation results illustrate that the mean distance error decreases as the number of access points and collected samples reduce.Secondly, in view of the complexity of constructing fingerprinting database, a new radio map construction method is proposed. This method divides the calibration points(CP) in the localization region into primary CPs and secondary CPs. The former are preselected and the number of which is limited and scattered in the experimental area. Then the CPs diffuse around on the basis of primary CPs and use 1m as the step size. Moreover, the log-distance path loss model is applied to calculating the locations of secondary CPs according to the internal relations of the triangle. This process repeats until the fingerprinting database is completely patched.Finally, the algorithm based on Voronoi diagram and Naive Bayes is used to locate the mobile targets. In offline phase, the RSS values are collected and processed by Gaussian filter, and the results of the mean values are stored in the fingerprinting database. Then the pre-selected CPs are used as the Voronoi generators to construct the Voronoi diagram, and each Voronoi region contains a primary CP and a plurality of secondary CPs. In online phase, we collect the real-time RSS values of the target and calculate the Voronoi region where the target is located. Furthermore, the naive Bayes algorithm is working to calculate the calibration point with the maximum posteriori probability restricted in the Voronoi region. This position is regarded as estimate location of mobile target. The experimental results show that the proposed algorithm is effective. |
PDF Full Text Request