Transformer Based Step Length Estimation Model for Waist-Mounted IMU

The Pedestrian Dead Reckoning (PDR) method, owing to its characteristics of being independent of external signals and possessing robust anti-interference capability, exhibits irreplaceable and significant application value in personnel positioning scenarios under complex environments such as emergency rescue, underground space navigation, and indoor robot scheduling. However, existing PDR technologies based on waist-mounted Inertial Measurement Units (IMUs) still face prominent step length estimation error issues in practical applications-affected by factors including fluctuations in waist motion posture, switching of multi-rate walking modes, and noise interference in inertial data, traditional step length estimation methods struggle to accurately capture the dynamic changes of gait time-series features, resulting in large step length prediction deviations. This problem largely restricts the overall positioning accuracy of PDR systems and has become a key bottleneck for their technical implementation in high-precision positioning scenarios. To address this technical bottleneck, this study designs a high-precision step length estimation model based on the Transformer architecture. The model innovatively integrates multi-source inertial data collected by waist-mounted IMUs. Through positional encoding tailored to the characteristics of gait data and multi-head self-attention mechanism, it enhances the deep mining of time-series features. Meanwhile, combined with multi-rate motion pattern analysis, it optimizes the feature extraction strategy for different walking states in a targeted manner, ultimately achieving accurate step length estimation under complex motion conditions. This research provides a feasible and efficient technical solution for high-precision personnel positioning in indoor environments, laying an important foundation for the engineering application and performance upgrading of PDR systems.