Search for a command to run...
Global navigation satellite systems (GNSS) can provide high-quality location information in outdoor environments. In indoor environments, GNSS cannot achieve accurate and stable location information due to the obstruction and attenuation of buildings together with the influence of multipath effects. Due to the rapid development of micro-electro-mechanical system (MEMS) sensors, today's smartphones are equipped with various low-cost and small-volume MEMS sensors. Therefore, it is of great significance to study indoor pedestrian positioning technology based on smartphones. In order to provide pedestrians with high-precision and reliable location information in indoor environments, we propose a pedestrian dead reckoning (PDR) method based on Transformer+TCN (temporal convolutional network). Firstly, we use IMU (inertial measurement unit)/PDR pre-integration to suppress the inertial navigation divergence. Secondly, we propose a step length estimation algorithm based on Transformer+TCN. The Transformer and TCN networks are superimposed to improve the ability to capture complex dependencies and improve the generalization and reliability of step length estimation. Finally, we propose factor graph optimization (FGO) models based on sliding windows (SW-FGO) to provide accurate posture, which use accelerometer (ACC)/gyroscope/magnetometer (MAG) data to establish factors. We designed a fusion positioning estimation test and a comparison test on step length estimation algorithm. The results show that the fusion method based on SW-FGO proposed by us improves the positioning accuracy by 29.68% compared with the traditional FGO algorithm, and the absolute position error of the step length estimation algorithm based on Transformer+TCN in pocket mode is mitigated by 42.15% compared with the LSTM algorithm. The step length estimation model error of Transformer+TCN is 1.61%, and the step length estimation accuracy is improved by 24.41%.