Research On Sampling And Matching Algorithms In WiFi Fingerprint Based Indoor Positioning System

With the rapid development of information technology, terminal-based indoor positioning technology has enabled various services, including information retrieval, indoor navigation, community dating etc. Existing indoor positioning methodologies can be categarized as: triangulation method, proximity method, WiFi fingerprint based positioning method and pedestrian dead reckoning. As WiFi fingerprint based positioning method has advantages of wide positioning range, low cost, flexible usage and no requirement of additional hardware, this thesis mainly deals with the key issues in this method.Basically, there still exists several problems in traditional WiFi fingerprint based positioning method. Firstly, in sampling phase, we should collect and preprocess signals at each sampling point, existing unidirectional collecting and mean filtering method is still not ideal. Secondly, in positioning phase, the matching accuracy of some typical existing algorithms such as KNN still need to be increased.In sampling phase of WiFi fingerprinting method, for the functionality of signal collecting, unidirectional collecting method is analyzed. Considering the fact that it has not considered the difference of signal strength when the phone points to different direction, so the method of collecting signal in different direction is proposed. For the functionality of signal preprocessing, the existing mean filtering algorithm is analyzed. Considering the weakpoint that it has taken into sum averaging some signals which deviate largely from mean value, so the method of Gaussian filtering is proposed to filter out these signals. Finally, an improved sampling algorithm----FODG(Fusion of Different direction collection and Gauss Filter) is proposed which integrates the method of collecting in different directions and Gaussian filtering.In positioning phase of WiFi fingerprinting method, the mainstream matching algorithm KNN is improved from the aspects of Euclidean distance calculating and coordinate matching. For Euclidean distance calculating process, the weakness is analyzed that KNN gives the same weight to each AP(Access Point), and the method of giving different weights to APs with different signal strength is proposed in this thesis. For coordinate matching process, the weakness is pointed out that KNN gives the same weight to the K neighbor reference points, and the distance weighted KNN method(WKNN) is proposed. Finally, an improved matching algorithm---- AWKNN(AP weighted and distanced weighted KNN) is proposed which integrates AP weighted Euclidean distance and WKNN.Finally, the whole WiFi fingerprint based indoor positioning system is designed and implemented on the Android platform, including sampling phase and positioning phase. Based on WiFi fingerprint based system, investigation of the optimal selection of relevant parameters(number of APs collected, number of WiFi signal collected and value of neighbors K etc.) is conducted. Then, the performance analysis of sampling algorithm FODG and matching algorithm AWKNN is carried on, and the whole positioning system(FODG + AWKNN) is tested from the aspects of positioning accuracy, positioning stability and positioning speed. The results show that the improved system can increase positioning accuracy and positioning stability to certain extent compared to traditional methods, and the positioning speed just slightly decreases.