Research On Indoor Fusion Positioning Algorithm Based On Neural Network

With the advent of the Internet of Things era,indoor positioning technology is widely used in people's lives.Reliable,real-time,and accurate indoor positioning technology has high application value and commercial prospects in scenarios such as product recommendation,fire rescue,and logistics transportation.The indoor positioning technology based on external sources requires additional equipment,which is costly and difficult to popularize.The indoor positioning technology based on built-in sources relies on inertial sensors and does not require additional equipment.It has strong autonomy.In view of the above,this paper combines data fusion,map matching and neural networks,and studies multi-source data fusion positioning algorithms based on generalized regression neural networks,and particle fusion neural network-based map fusion positioning algorithms.In this paper,multi-source data fusion positioning is firstly studied.Aiming at the problem of poor positioning effect in multi-device sensor data fusion positioning,a GRNNAdaBoost algorithm is proposed.The algorithm is based on a generalized regression neural network model,and uses interactive integrated learning methods to fuse the built-in sensor data of different devices to better infer the indoor position of pedestrians.For the sensor data collected from a smartphone and a smart watch at the same time in a real scenario,the GRNN-AdaBoost algorithm is used to use the correlation between sensor data collected by multiple devices at the same time to perform position prediction.After experimental verification,the GRNN-AdaBoost algorithm has a positioning error of 1.09 m and0.83 m in the two directions of the two-dimensional map,which is superior to the CART regression tree and the AdaBoost algorithm.In this paper,the map fusion positioning is studied.Aiming at the problem of lack of feedback mechanism in map fusion positioning,a map fusion positioning algorithm based on particle filter neural network is proposed.Based on the recurrent neural network as the basic framework,the overall differentiability of the particle filter neural network is realized.In order to better extract map information,perform local segmentation on the map,and use convolutional neural networks to learn map features,improving the accuracy of particle weight discrimination.When the local map pixel size is 80 × 80,the convolutional neural network The accuracy rate can reach 98 %.Experimental analysis of the algorithm performance in empty rooms,narrow corridors and free areas respectively shows that the proposed algorithm is better than the traditional particle filtering based map fusion positioning algorithm.This paper uses neural networks for data fusion and map matching.The neural network is trained using inertial navigation data collected in real scenarios.The multi-source data fusion positioning algorithm based on generalized regression neural network and the map based on particle filter neural network are verified.Good performance of fusion positioning algorithm.