Inertial Navigation With Adaptive Sliding Window and Spatiotemporal Network

The performance of pedestrian inertial navigation systems (INS) is often compromised by frequent directional changes, sharp turning angles, and complex walking trajectories at varying motion speeds. Conventional INS frequently struggle with error accumulation and sensor noise under such conditions. To overcome these limitations, this study introduces the adaptive sliding window and spatiotemporal network (ASWINS). The adaptive sliding window mechanism categorizes pedestrian motion into turning and linear segments. This segmentation enables precise directional analysis for subsequent refined localization. The network further incorporates a spatiotemporal attention module to extract features from segmented inertial measurement unit (IMU) data, which enhances motion dynamics analysis and improves velocity estimation accuracy. Through comprehensive testing across diverse real-world environments, ASWINS demonstrates significant improvements in position estimation accuracy and robustness across different walking states and environments. Moreover, ASWINS reaches state-of-the-art (SOTA) levels in these scenarios, which confirms its consistent advantages in handling intricate navigation scenarios characterized by rapid directional shifts and speed variations.