Research On Improved WLAN Position Fingerprint Localization Algorithm Based On Chi-square Distance

Recently indoor localization technology is anticipated to play an increasingly important role in the modern society. In general, indoor localization algorithm can be categorized into two main classes:parametric localization algorithm and non-parametric localization algorithm. In all these algorithms, position fingerprint localization algorithm which belongs to non-parametric localization algorithm has been studied and applied extensively. Position fingerprint localization algorithm consists of two phases, the offline training phase and the online localization phase. In the offline training phase, we collect a set of RSS values at each RP from various APs to form one of the fingerprints and store it in the database. In the online localization phase, we compare the real-time fingerprint collected by the user mobile terminal with the fingerprints stored in the database to obtain the user physical coordinates. This thesis focuses on RSS-based position fingerprint localization algorithm in the WLAN environment and mainly explains the methods for RP clustering and AP weighting.Through analyzing the widely used K-means RP clustering algorithm and aiming at the fact that it is very sensitive to noise and outliers, this thesis proposes using Fast K-medoids algorithm to cluster the RP. In the offline phase, the Fast K-medoids RP clustering algorithm selects initial medoids among the RP fingerprints first, then finds the medoid of each cluster, which is the fingerprint minimizing the total distance to other fingerprints in its cluster by using iterative method. Compared with the K-means RP clustering algorithm, Fast K-medoids RP clustering algorithm is more robust against noise and outliers.In position fingerprint localization techniques, WKNN is a classic pattern matching algorithm. Euclidean distance is usually used as the measure function of the WKNN algorithm, it assigns the same weight of different AP in the localization area, but different AP has different influence on the localization accuracy. For this problem this thesis proposes adopting Chi-square distance that can more reflect the relative relationship between the feature vectors and using sensitivity method to compute the AP weights. Then we use weighted Chi-square distance to calculate the user coordinates. This algorithm can be called the improved WKNN localization algorithm based on Chi-square distance. It can reduce the effects of environmental noise. Furthermore, we can combine this method with the Fast K-medoids RP clustering method to form a complete algorithm. It can be called the improved position fingerprint algorithm based on Chi-square distance.The experiment indicates that the improved WKNN localization algorithm based on Chi-square distance has greater localization accuracy than WKNN algorithm. By combining it with the Fast K-medoids RP clustering algorithm, we get the the improved position fingerprint localization algorithm based on Chi-square distance. This algorithm can reduce the complexity of WKNN localization algorithm while improving its localization accuracy. The proposed algorithm provides an important reference for the realization of efficient indoor localization.