Research On Fusion Localization Method For Multi-target Network Without Anchor

The increasing popularity of multi-target systems,such as multi-mobile robot systems and clustering systems,highlights the need for highly reliable Multi-target Localization(MTL).However,most MTL systems currently rely on base stations and other infrastructure,which limits their practical application range to some extent.Therefore,researching MTL systems that do not require base stations is of great significance.Among the various technologies available for Multi-target Localization(MTL),odometer and ranging technology are widely used due to their maturity and low cost.However,odometer has the problem of cumulative error,and in the absence of a base station,the distance measurement information between individuals can only provide the relative topological structure of the multi-target position,but not the position information.By integrating these two types of information,the error caused by the odometer can be reduced using distance constraint,and the problem caused by only the ranging can be simultaneously addressed.This is an effective approach to solve the MTL problem without a base station.Currently,fusion methods can be broadly categorized into three types: those based on Extended Kalman Filter(EKF),probability estimation,and optimization.However,the EKF-based method has a large linearization error and the filter is prone to divergence.Probabilistic statistics-based methods rely on many prior conditions.Generally,the optimization-based method directly uses the measured distance to establish the optimization equation,which results in a larger optimization problem and a higher probability of reaching local optima.Additionally,the typical optimization method only utilizes distance information related to the individual,and the utilization rate of ranging information and positioning performance requires improvement.This thesis proposes an optimization-based method that utilizes the topological structure obtained from ranging to fit the prior estimation results obtained by the odometer to obtain the target position estimation.The proposed method maximizes the use of ranging information,which significantly improves Multi-target Localization(MTL)performance without a base station.In addition,the thesis presents corresponding versions of the MTL algorithm for three common multi-target positioning scenarios.The main innovation points and key problems addressed in this thesis are as follows:(1)A Multi-target Dead Reckoning(MDR)algorithm is proposed.By using MultiDimensional Scaling,the MDS algorithm converts the topology structure of multitarget location based on ranging into the form of relative coordinates.Then,the optimization problem is converted into the Procrustes problem about the location result of the odometer and the relative topology structure of multi-target position.Finally,the closed solution of the optimization problem is obtained by the proposed topology fitting method.This method not only simplifies the original optimization problem but also improves the utilization rate of ranging information and solves the problem of multitarget 2D positioning without a base station in the absence of prior information.(2)An Adaptive Multi-Target Dead Reckoning(AMDR)algorithm is proposed.Based on the optimization problems established by MDR,weight factors are added to this model,and the optimization problems are solved by an adaptive topology fitting method.Then,the analytic expression of the weight factor is derived based on the minimum variance estimation criterion.Based on inheriting the performance of MDR,this model solves the problem of multi-target location without a base station in a 2D condition with known prior information of the noise model.(3)A 3D Adaptive Multi-target Dead Reckoning algorithm(AMDR3D)is proposed.The proposed method extends the MDR and AMDR models respectively.When solving the 3D topology fitting problem,we use the Singular Value Decomposition(SVD)method in the 3D rigid body rotation to get the optimal solution of the third-order rotation matrix.Finally,the analytic expression of all node position estimation is obtained.The problem of multi-target location without a base station is solved in a 3D condition when there is prior information and there is no prior information.The proposed MTL model adopts the loose coupling fusion mode,the framework is more general,has high scalability,and is suitable for most odometer and ranging technologies.At the same time,the simulation results show that the error of the odometer corrected by MDR is about 70% under the assumed experimental conditions,and the positioning performance of MDR is improved by about 50% compared with the general optimization-based method.The simulation experiments of AMDR model and MDR model under different ranging conditions show that the error of the AMDR algorithm is greatly reduced in the initial stage compared with MDR,and the ability of ranging to correct the cumulative error is improved when the ranging error increases.The simulation experiment of AMDR3 D shows that the performance of the AMDR3 D model is almost the same as that of the 2D model.