Research On Indoor Positioning Method Based On Adaboost

In recent years,the improvement of technology has promoted the changes in lifestyles,and location-based services have become a very important part of life.In outdoor environments,satellite navigation systems have been perfected.In comparison,the services are very immature in indoor environments.Therefore,positioning services in indoor environments have become a important research direction.Wireless local area network-based indoor positioning technology has attracted the attention of researchers due to its low hardware costs,flexible networking methods and convenient infrastructure.The initial researchs focus on the characteristics of Received Signal Strength Indication(RSSI),but with the continuous development of hardware technology,researchers have found that relying on RSSI alone will lose a large amount of multipath information.The characteristics based on Channel State Information(CSI)can better reflect the changes in the environment.Therefore,this dissertation proposes a passive positioning method based on CSI and use ensemble learning to train fingerprint models.The main work is as follows:(1)After analyzing the signal propagation model and positioning advantages in the Wireless Fidelity(WIFI)environment,CSI was selected as the characteristic fingerprint.By collecting CSI data over a period of time for experiments,it is verified that CSI has the characteristics of time stability,frequency diversity and spatial difference,which provides a theoretical basis for subsequent experiments.(2)This dissertation uses density-based clustering to process the data collected in the experiment.Through the judgment of the spatial distance,the edge data in the fingerprint set is selected as anomalous data,and it is eliminated to ensure the reliability of the experimental data.Principal component analysis is used to reduce feature dimension and the dimensionality-reduced information is used as the input fingerprint feature.(3)Select adaboost,a kind of ensemble classifiers,as the main classifier to improve the accuracy of learning through multiple learning.Finally,the output of the classifier is transformed into two-dimensional coordinates of the space through confidence probability regression.Turning positioning into a regression problem finally.After experiments in two representative environments,an empty corridor and a complex laboratory,it is found that the passive positioning method proposed in this dissertation can effectively improve the accuracy of indoor positioning and achieve the expected results.The average error in an open corridor can reach 0.88 m while the distance of fingerprint is 1m,and the error in the laboratory will increase by 0.5m compared to the corridor.Compared to other regression methods,the positioning accuracy using probability regression is greatly improved.