|
Kernel fuzzy C-means(KFCM) algorithm is proposed to manage fingerprint database (also known as radio map) collected from WiFi indoor positioning system. Clustering method greatly reduces the calculation load of position match process. By setting the cluster amount k before initialization, after constant iteration of cluster centers' renewal as well as membership' update until the objective function is minimized or reaches the preset conditions, the database is fuzzily divided into k clusters with k representative centers. With the purpose of obtaining superior clustering performance and better localization precision, this paper proposed several improvements. First, choose optimal interval of reference points(RP) and the amount of access point(AP) to build fingerprint database in order to enhance positioning precision; second, deduce an applicable cluster amount based on the structure characteristics of fingerprint database using sample density method; third, by approximating actual kernel matrix to a hypothetical ideal kernel matrix derives to a kernel parameter that is more appropriate for the radio map. Assessment criteria VXB is proposed to verify the effectiveness of kernel parameter's optimization based on Iris dataset. By means of the parameters adjustment above, the clustering result is more suitable for the selected datasets, and the last two improvements lead to 19.5% and 23.55% improvement in localization accuracy. |
|
Keywords:WIFI Localization, radio map, parameter optimization, kernel fuzzy C-means, clustering |
|