Research On Multi-Source Singal Of Opportunity Fusion Positioning Method

With the development of technology,high standard location based services(LBS)have become an indispensable information service,which is widely used in social life such as emergency communication,public transportation,intelligent medical care and big data development.in every way.The Global Navigation Satellite System(GNSS)has become the most widely used navigation means.However,since GNSS signals are easily blocked or absorbed by ground obstacles,especially in indoor and urban dense areas,GNSS cannot guarantee Accurate navigation services,so high-precision and stable indoor positioning technology is urgently needed to meet the growing demand for LBS.Navigation via signal of opportunity(NAVSOP)achieves user positioning by receiving and processing existing wireless signals and some non-wireless signals in space.With its various advantages,it has received extensive attention in the academic community in recent years.In order to meet the requirements of high-precision and reliable indoor positioning,this paper focuses on the problems of limited positioning accuracy and low robustness of a single signal of opportunity,focusing on Bluetooth wireless signal,geomagnetic signal of opportunity and pedestrian position estimation based on unscented particle filtering.Reckoning,PDR)multi-source fusion positioning method,proposed an extended fingerprint positioning method based on LightGBM and an improved unscented particle filter algorithm,and designed a multi-source fusion positioning system based on unscented particle filtering,which can effectively improve the positioning accuracy and Location robustness.The specific research work is as follows:Firstly,to solve the problem of noise and outliers affecting fingerprint data in indoor positioning,this paper focuses on an indoor positioning model based on LightGBM,which uses Bluetooth signal of opportunity and geomagnetic signal of opportunity to achieve extended fingerprint positioning.By collecting Bluetooth RSSI and geomagnetic signal strength,and performing normalization processing and Gaussian curve fitting respectively,an extended fingerprint of Bluetooth signal and geomagnetic signal is constructed,and the data is trained by LightGBM.Experiments show this method effectively improves the positioning accuracy and reduces the influence of factors such as multipath and noise interference,and increased computational efficiency.Secondly,to solve the problem that the state covariance of the sampled particles is not positive definite after multiple cyclic updates of the unscented particle filter,which leads to the inability to continue iterative updating or the filter results do not converge,this paper proposes an improved unscented particle filter algorithm,which applies singular value decomposition to In the UPF algorithm,the Cholesky decomposition of the sampled particle state covariance matrix and the inverse matrix of the observational covariance matrix are replaced,which reduces the indefiniteness of the sampled particle covariance matrix caused by the cyclic update of the UPF,causing the filter update to fail,which effectively improves the robustness of UPF algorithm.Finally,aiming at the problems of limited accuracy and low robustness of single signal of opportunity positioning,a multi-source signal of opportunity fusion localization system based on improved UPF is designed,which combines the positioning results of LightGBM model and PDR.Localization results of LightGBM as observations in the above improved UPF,and the PDR motion equation is used as the state update equation of the system.Experiments show the MSE of the multi-source fusion positioning system proposed in this paper is 0.74m,which is 48%higher than that of the particle filter.