The Improvement of Indoor Localization Precision through Partial Least Square(PLS) and Swarm(PSO) Methods

• Main Authors: LIANG,CHAO-LUNG, 梁昭隆
• Other Authors: SHIH,CHIH-HSIONG
• Format: Others
• Language: zh-TW
• Published: 2016
• Online Access: http://ndltd.ncl.edu.tw/handle/75gyvf

碩士 === 東海大學 === 資訊工程學系 === 104 === In recent years, indoor location technology has undergone a period of rapid development. Many methods have been proposed and used including Time of Arrival(TOA), Time Difference of Arrival(TDOA), Arrival of Angle(AOA), Ultra Sonic, Received Signal Strength Indicator (RSSI). They are carried out using a wireless signal for indoor positioning needs. Fingerprint positioning method collects the locational characteristics to establish fingerprint database in specific forms. After receiving the localization request from users, comparison on the database is exercised for positioning purposes. Gaussian model is used to build the fingerprint database based on RSSI values. After receiving the positioning request, an anchor point.is searched in the fingerprint database based on a maximum probability match. The single Gaussian model is very simple to approximate database, but the probability distribution of the original data is difficult to comply with. On the other hand, the use of Gaussian mixture model can reduce the error value, but it requires a lot of computing. A PSO algorithm is used to enhance the approximate accuracy within a reasonable extent. However, there are shortcomings that approximation converges.easily to local extremes prematurely. Therefore, we slightly improved to avoid falling into local extreme conditions, the total amount of the original data error probability distribution is 0.1 to 0.5 times from the original. In order to obtain more accurate location result, we try using the partial least squares(PLS) model to establish fingerprint database. Coupling genetic algorithm-based partial least squares(GA-PLS), we are able to reduce sampling point positioning calculation to not only enhance localization error but improve the computational efficiency of positioning. Using PLS, the positioning accuracy reaches up to 91%, with the average amount of error reduced to 15.5 cm. It is better then Gaussian model and Gaussian mixture model. The single Gaussian model has a positioning accuracy of 44%, with the average amount of deviation error of 94 cm. Using Gaussian mixture model, the positioning accuracy is 65.15%, with the average amount of error is 75.5 cm.