The Optimization Of Fingerprint Technology In WLAN-Based Indoor Localization

With the wide popularity of intelligent terminals, mobile terminal-based applications are becoming diversified. Thanks to the diversity of intelligent terminals, the indoor localization system gets rid of the shackle which requires to deploy additional hardware. But the new problem appears, the RSSI(Received Signal Strength Indication) measurements might be different as the developers and users have different terminals in the WLAN fingerprint-based positioning systems. In addition, due to the natural or man-made reasons, the reference AP(Access Point) cannot be accessed or location has been changed, which will result in the poor positioning. Calibration methods for combat these problems, such as linear(or nonlinear) mapping algorithms and BP neural network algorithms, require offline training before usage. In order to mitigate the impacts due to the diversity of mobile terminals and the anomalies of reference AP,this thesis analyzes these problems in the finger-print positioning systems. Accordingly, the work of this thesis includes the following sections:First of all, for the difference of RSSI measurements arising from the diversity of mobile terminals, this thesis proposes an indoor location algorithm based on the Pearson correlation coefficient, by analyzing the similarity of curve between the fingerprint and the measurement by users. Compared with the linear and nonlinear mapping algorithms, it does not need training in advance, and reestablishing the model when the environment changes. The pearson-based method is compared to the Euclidean distance-based location algorithm on the same environment. Results show that the locating error in the range of 2 meter decreases by 8% for the pearson-based method.Then, by analyzing the un-accessing or movement of reference AP, this thesis proposes the related solutions. There were two parts: firstly,to the un-accessing of AP, this thesis proposes a weight-based localization algorithm by calculating the number of AP in the intersection of the fingerprint and user data, defining the weight factor, and calibrating the positioning result. Results show that the location accuracy increases 9% in the range of 3 meter. Secondly, for the location changing or environment of reference AP, this thesis judges the change by combining the pearson coefficient with crowdsourcing. When the position or the surrounding environment of reference AP changes, the measurement is significantly far from the data of the fingerprint database in the same position. According to this feature, this thesis defines the location related pearson coefficient with crowdsourcing to judge whether the AP is moved.In the end, this thesis compares the positioning accuracy of pearson-based algorithm with the Euclidean distance-based algorithm based on the fingerprint database which is built with lots of intelligent terminals. Simulating the procedure which the crowdsourcing builds the fingerprint database, the values of RSSI are measured in the same environment, the same position, and with multiple mobile terminals, to build the fingerprint database. Experimental results show that pearsonbased algorithm can improve the locating accuracy about 30% in the range of 2 meter compared to the classical Euclidean distance-based algorithm.