An Indoor Positioning Method Based On UWB And Visual Fusion

Continuous localization and tracking of multi-pedestrian objects is a common concern in many applications such as large indoor space security,emergency evacuation and location services.Among various positioning methods,Ultrawideband(UWB)is an important technology to achieve high precision indoor positioning.However,due to the presence of indoor non-line-of-sight(NLOS)errors,a single positioning system can no longer meet the requirements of high precision positioning.Based on the integrated indoor positioning method of UWB and vision,this paper focuses on the following research on positioning accuracy and stability:(1)Aiming at the problem that the traditional UWB positioning solution method does not make full use of the feature information of the positioning data in the time series,a positioning solution method based on the Gated Recurrent Unit(GRU)network model is proposed.Different from the optimization direction of traditional positioning algorithm,the bidirectional GRU model has less parameter Settings and high efficiency.It can learn the data features of historical time series and future time series and modify the model.The data features of future time series are closely related to the orientation information in the positioning process,so compared with the oneway GRU network model,The bidirectional GRU network model provides deeper information for location prediction.Experimental results show that the UWB positioning accuracy based on the bidirectional GRU algorithm is improved by 19.23%and 4.9% compared with Chan-Taylor and one-way GRU algorithms,respectively.(2)Aiming at the problems of low calibration accuracy and poor robustness due to lens distortion and measurement error during camera calibration,a monocular camera calibration algorithm based on improved particle swarm filtering is proposed.On the basis of the classical particle swarm optimization algorithm,the idea of population co-evolution is borrowed,and the global suboptimal factor is added to the updating equation,which improves the robustness and convergence of the algorithm.Then,in the calibration process,the camera reference value calculated by Zhang’s camera calibration is used as the reference value of the initial range of particles,which is input into the improved particle swarm optimization algorithm for further iterative optimization to improve the accuracy of camera calibration.Experimental results show that the improved algorithm has higher calibration accuracy and faster convergence speed,and effectively improves the problem that particles are easy to fall into local optimal solutions.Finally,a security detection software is designed and developed by using the visual positioning method proposed in this paper.(3)Aiming at the problems of the NLOS error of single UWB positioning,the blind area influence of single visual positioning and the uncertainty of pedestrian identity,an indoor positioning method based on UWB and visual fusion is proposed.Firstly,the pedestrian ID and pixel coordinates are obtained by the visual object detection and tracking algorithm,and the pedestrian world coordinates are calculated by the coordinate transformation model.At the same time,the pedestrian holds the UWB tag,and the signal obtained by the UWB base station is transmitted to the server for positioning and calculation.Then,with Euclidean distance as the cost matrix,the Hungarian algorithm is used to match the identity between UWB and visual positioning results.After successful matching,the federated Kalman filter algorithm is used to fuse the results of the two methods.Finally,the experiment shows that the fusion positioning accuracy is 22.43 cm,and the average positioning error is reduced by 32.54% and 25.63% compared with the single UWB and visual positioning,respectively.