Wi-Fi And Bluetooth Fusion Based Indoor Localization Algorithm And System

As the indoor activities space of people is becoming more and more tremendous and complex,people's point of interest has become more rich compared to before,and the requirements of localization and navigation change into Increasingly strong in parking lots,supermarkets,airports and etc.What's more,it is requisite for some industries such as precise location marketing,intelligent manufacturing,robots and unmanned medical care to identify location of special objects in the indoor environment.What we have mentioned above provides huge opportunities for indoor positioning system.Statistics indicate that people who spend time indoors accounted for above 80%,and there have become increasingly difficult to contact relatives and friends,go shopping malls to find specific goods and find their cars in parking lot because of the increasingly complex and huge indoor environment.Currently,indoor positioning technology showed a hundred schools of thought contend.However,there is no indoor positioning technology that can satisfy localization requirements for location based service with low cost,and ultra-wideband positioning technology,laser positioning,infrared positioning and geomagnetic positioning are difficult to popularize due to specialized equipments or complex deployment methods and high cost.But Wi-Fi based fingerprint positioning technology is easy and sixpenny to deploy by taking use of existing equipment,and bluetooth low energy based localization technology is in the ascendant and full of hope with low energy,extensive coverage of signals and low cost.However,there are some limitations for Wi-Fi and bluetooth low energy single mode localization methods: Firstly,fingerprint based methods are difficult to applied in the practical applications because of the refection,diffraction and multipath effect of signals in the complex environment;Secondly,for most of traditional approaches,the location estimation problems assume the availability of a vast amount of labeled calibrated data,which requires a great deal of manual effort and long training time on offline phase.To solve these problems mentioned above,we commence them in stability of fingerprint features,speedability of model training,convenience of fingerprint acquisition and localization accuracy,and the main work in this full text can be divided into the following three parts:1)We propose a cross-correlation based fusion feature extraction method.Firstly,we apply gaussian model to denoise original sensor data,then fusion features are calculated according to cross-correlation and combined with original sensor features.Experimental results show that fusion feature we proposed can improve the stability of fingerprint feature and accuracy and robustness of positioning model.2)We propose a semi-supervised localization method basing on fusing feature.First of all,we employ extreme learning machine to promote the training speed and generalization ability of model.Secondly,a semi-supervised learning method and Laplacian regularization are employed to import large number of unlabeled samples which can dramatically reduce labeled calibration samples for profiting from unlabeled samples and reducing the negative impact.Comparative experiments show that the semi-supervised manifold localization method can dramatically reduce labeled calibration samples by 90% and increase the indoor localization accuracy by 20%-30%.3)We design and realize an indoor fusion positioning system.This system provides a service which helps Third-party developers embed positioning function quickly in their mobile applications.Third-party developers just calibrate some fingerprint data by using our fingerprint collection tool and upload them to our cloud platform,then them can use offline localization SDK to experience indoor positioning function quickly.The system test and trace recurrence experiment show that our indoor positioning system is easy to use and has high commercial value.