Research On Indoor Location Algorithm Of LSTM Neural Network Based On Smart Phone

With the development of the mobile Internet era,people's demand for their own positioning is increasing,and the demand for location no longer stays outdoors,and indoor positioning quickly enters people's field of vision.However,because Global Position System(GPS)signals are easily blocked in indoor environments,indoor positioning algorithms have been unable to achieve the accuracy of outdoor positioning algorithms.At present,most indoor positioning technologies have common shortcomings.They need to deploy their corresponding specific equipment in advance before positioning.The cost is high and there is spatial locality,which cannot be implemented in daily life.The inertial navigation system relies on its own built-in sensors to obtain position-related information such as speed and attitude angle to achieve positioning.It is an autonomous navigation system that is not restricted by time,space,etc.,has low cost of ownership,low power consumption,and high performance.characteristic.With the development of microelectronics technology,more and more high-performance sensors are integrated into smart phones,so indoor positioning based on smart phones has a widerangeof application prospects.First of all,for pedestrians in the indoor environment,there are fewer changes in their motion states.This paper uses the quaternion method to update the attitude.In order to avoid the interference of the complex ground environment on the positioning accuracy,a pedestrian motion model based on curvature is established.Secondly,in view of the problem of Gaussian white noise in the output data affected by the surrounding environment during the acquisition of low-cost MEMS sensors on smart phones,Kalman filter is used in MATLAB to preprocess the gyroscope sensor data to eliminate Gaussian white noise and reduce Impact on positioning accuracy.The neural network can train the model without prior knowledge,and use the validation set data to verify the trained model.In view of the interference from the constant drift of the low frequency band,because the gyroscope data belongs to the time series data,the traditional neural network cannot consider the time sequence of the data,so this article introduces the long short term memory artificial neural network(Long Short Term Memory,LSTM)In order to reduce the singular sample data,the gyroscope data is normalized.In order to improve the model stability and prediction accuracy,a sliding window with a label length of5 is selected to construct the data set.Seven different experimental schemes are designed to determine the main parameters of the final LSTM model,Adam optimization algorithm is selected to adjust the network weights,data prediction is performed through LSTM,drift errors areeliminated,and accumulated errors are furtherreduced.Experimental verification shows that compared to directly using MEMS sensor data for positioning,the indoor positioning method based on LSTM can significantly improve the positioning accuracy,with an average error of 1.33 meters,which meets the needs of people's location services.