Research Of Fusion Indoor Localization Algorithm Based On UWB Ranging Optimization And Image Target Detection

Location service is widely used in all aspects of life and indispensable in every field as society becomes more and more intelligent and informationalized.The demand for indoor location service is gradually increasing in the complex indoor environment,so the highprecision,stable and reliable indoor localization technology has become a hot research topic at present.Among the existing indoor localization technologies,UWB-based localization technology has attracted much attention due to its fine security and high localization accuracy,but it is easily interfered by complex indoor environmental factors.Instead,image localization technology is less affected by wireless signals and low-cost,but its localization accuracy is not guaranteed.However,multi-source fusion localization can make up for the defects of single localization source and realize complementary advantages,and it has become a research trend in the field of indoor localization.In view of the limitations of single localization technology and the advantages of multi-source fusion localization,this thesis carried out research on fusion indoor localization algorithm based on UWB ranging error optimization and image target detection.The main research work is as follows:Firstly,we propose a UWB indoor ranging error optimization algorithm based on CNN.Aiming at the localization error elimination of UWB system in non-line-of-sight environment,three neural network models are selected to optimize the error.The self-defined data set is used to train and test the network,and the optimal network model was screened out by comparing the experimental results,which effectively reduced the distance error of UWB localization.In network training,a gradient descent optimization method based on gradient direction is proposed and its feasibility is proved.Secondly,we propose a double neural network indoor localization algorithm based on image target detection.In this algorithm,YOLOV5 with coordinate attention mechanism is used to identify the locator information,then the localization network is used to convert the detection information into location information to achieve location estimation.The self-defined data set is used to train and test the network,the results are compared with that of the existing image localization algorithms.The experimental results indicate that the indoor localization algorithm used double neural network based on image target detection improves the accuracy and stability of image localization.Finally,we propose a fusion localization algorithm based on Unscented Kalman Filter.In this algorithm,the localization data after UWB ranging error is combined with the localization data based on image target detection by Unscented Kalman Filter algorithm,and the fusion results are further fitted and optimized by the neural network basic perceptron to obtain more accurate localization results.Experimental results show that the data fusion localization algorithm based on Unscented Kalman Filter performs better than the single UWB or image localization.