An integrated model using implicit constraint generator, fingerprint based simulator and multi objective optimization for indoor localization of moving object

Abstract

Indoor localization is one of the most active research topics. It involves utilization of various sensing and technologies to accomplish global positioning system alternative solution for indoor localization. WiFi based indoor localization is regarded as the most promising sensing technology for non-invasive indoor localization with adequate accuracy. Typical approach of building WiFi based indoor localization is WiFi fingerprinting based on site survey and training machine learning to predict the location. This can be applied to any moving object as long as it is occupied with WiFi sensing and to any moving pedestrian accompanied with smartphone or tablets. Despite the significant research for developing WiFi based indoor localization, the literature is still yet to resolve various issues. Most importantly, multi-path and jumping behaviour, the dynamic aspect of navigation runs and topology optimization. This thesis tackles these aspects and aims at resolving them by proposing a model for WiFi based localization with an integration of various components. First, it develops a model for improving the accuracy of indoor localization based on the implicit constraints. Second, it develops an algorithm that simulates navigation behaviour in indoor navigation environment and use it for converting fingerprint to sequential navigation data. Third, it integrates implicit constraints model with online learning classifier for predicting the location based on fingerprint. Fourth, it develops a multi-objective optimization algorithm based on introducing the concept of crowding angle to optimize localization classifiers and integrate it with the localization algorithm. The components of the model and the overall model were evaluated using state of the art approaches and benchmarks. The evaluation has included the verification of the superiority of the developed multi-objective optimization algorithm in the exploration and optimality. Furthermore, the present research evaluates the developed model with all its components generated navigation runs. Also, comparisons of its accuracy with both online sequential extreme learning machine (OSELM) and feature adaptive OSELM are conducted. Accomplished accuracy of the present model is around 95% with superiority over the benchmarks.