Smartphone Heading Correction Method Based on LSTM Neural Network

The error of heading estimation is the main factor that affects the accuracy of the navigation and positioning of the smartphone based on the pedestrian dead reckoning. In this paper, aiming at the heading error caused by gyroscope drift in smartphone pedestrian heading estimation, a smartphone heading error compensation model based on Long Short-Term Memory network (LSTM) is proposed. Due to the low precision of inertial devices in smartphones, the direct integration will lead to cumulative errors that the neural network can correct. Considering that the calculation method of inertial navigation is dead reckoning, the data of the preceding and following moments is related. And LSTM performs better in the long sequence than the ordinary neural network, so this paper uses a motion capture system to get high-precision attitude information. The information is used as true value to establish the heading error correction model of smartphones. By comparing the proposed algorithm with traditional quaternion heading estimation method, the error correction model presented in this paper can effectively suppress the influence of gyroscope drift. The result is better, which can meet the basic needs of pedestrian navigation and positioning.