Research On Indoor Positioning Technology Without Infrastructure Based On Acoustic Fingerprints

Mobile Internet has penetrated into our lives and provided us with various location services,but traditional satellite navigation can only provide reliable services in an outdoor environment,and people spend most of their day in an indoor environment,so that the market demand for reliable,accurate,low-cost indoor location services continues to grow.There are all kinds of sounds in the natural environment,we can obtain these sounds containing rich information anytime and anywhere,and there must be some differences between the same acoustic characteristics of sounds obtained from different scenes or different areas.Through the analysis of acoustic features by computer,the source of sound can be determined.According to this method,the computer can realize the positioning function without any infrastructure.Before the writing of this paper,the previous researchers have done a lot of study on the indoor location scheme based on acoustic fingerprint.Nevertheless,there are still some challenges to be solved in the real world of these studies:the acoustic fingerprint data extracted from indoor ambient sound is highly redundant and takes up a lot of computer resources,as well as the update of the acoustic fingerprint database requires time-consuming manual site survey,resulting in high cost of the system.In view of the above problems,this paper has made the following research work and contributions:(1)The extraction methods of three acoustic features Sonogram,Spectrogram and MFCC are studied.In addition,ambient sound data are collected from several indoor areas in the real environment,and the three acoustic features Sonogram and Spectrogram,MFCC are extracted,and two of them are combined to construct the acoustic fingerprint database,respectively.Finally,the classification experiment of KNN is carried out to test the discriminative ability of several combinations of acoustic fingerprints to the subarea of the room.The experimental results show that the combination of Sonogram and Spectrogram has the strongest discriminative ability.(2)The theory of principal component analysis(Principal Component Analysis,PCA)is studied and deduced,and the two acoustic features Sonogram and Spectrogram are optimized by principal component analysis,then the redundant data in these two feature matrices are removed,and improved acoustic features PCA-Sonogram and PCASpectrogram,are obtained.finally,the recognition accuracy of KNN to the original fingerprint combination and the optimized acoustic fingerprint combination is compared through experiments,and the experimental results show that the improved fingerprint combination has stronger discriminative ability.The average recognition accuracy of KNN classifier is improved by about 12% at most.(3)this paper also studies the algorithm principle of BP neural network and RBF neural network,and tries to generate some fingerprint data through the machine learning model to reduce the workload of updating fingerprint database.In the experiment,two kinds of acoustic fingerprints,PCA-Sonogram and PCA-Spectrogram,are used to train the machine learning model.The experimental results show that the fingerprint data generated by RBF neural network is closer to the real data.Finally,the classification results of KNN classifier are compared when using all real fingerprint data to construct database and using partial real data with data generated by RBF model to construct fingerprint database.The results show that the classification accuracy of the latter is very close to that of the former,which not only improves the updating efficiency of the fingerprint database,but also ensures the precision of the fingerprint database.