Research On Indoor Localization Using Multi-source Information Fusion And Machine Learning

With the rapid development of Internet of Things application,indoor localization has received more and more attention.Due to the complexity of the indoor environment,the existing satellite positioning techniques cannot meet high-precision indoor localization requirements.In order to solve the above problem,this thesis studied new indoor localization algorithms based on multi-source information fusion and machine learning.The main work can be summarized as follows:(1)The theoretical principle of indoor localization technology is studied.First,the ranging based radio positioning technology is introduced.Then the fingerprint matching algorithm and multi-source fusion theory are described which gives a theoretical foundation for the following research.(2)A multi-source localization algorithm based on the Laplace pyramid and pixel level fusion is proposed.First,in the offline phase,after the pre-processing,the received signal strength(RSSI)of mobile phone signal and bluetooth signal are formed as a coarse localization fingerprint by data splicing.The camera image decomposed by the Laplace pyramid and the Wi Fi image formed by RSSI measurements are transformed into a refined localization fingerprint through pixel fusion.The supporting vector machine(SVM)and the convolutional neural network(CNN)are used for coarse localization training and refined training respectively.At last,the coarse localization classification model and the refined localization regression model are obtained.Through the coarse and refined localization process,the offline learning efficiency can be improved.The experimental results show that the proposed algorithm has better performance than the existing localization algorithms.(3)A multi-source localization algorithm based on polynomial feature extraction and feature level fusion is proposed.First,in the offline phase,the dimension of the RSSI measurement of the mobile phone signal and bluetooth signal are expanded by the polynomial feature extraction.And then the two feature vectors are spliced to a coarse localization fingerprint.Meanwhile,the fully connected network and CNN are used to extract features from RSSI measurements of Wi Fi signals and the camera image of mobile phone respectively.The feature level fusion is used to form refined localization fingerprint.Then Ada Boost algorithm and fully connected network are used for coarse localization and refined localization learning,respectively.At last,the coarse localization classification model and refined localization regression model are obtained.In the proposed algorithm,the feature dimension expansion can get more signal features.Moreover,the feature level fusion has great flexibility.The experimental results verified the performance of the proposed algorithm.