A Robust Multi-IMU Sensor Fusion Based Lower Limb Evaluation Network Toward Rehabilitation

Wearable inertial sensors enable continuous and accurate monitoring of movement posture during rehabilitation training. However, challenges remain in posture estimation accuracy due to sensor placement bias, environmental interference, and the complexity introduced by nonlinear calibration models and multi-sensor data fusion. In this study, our previously proposed inclination-based linear magnetometer calibration model was validated through laboratory experiments involving dynamic indoor movements. Results demonstrate that the model effectively calibrates magnetometers without axes aligned, as verified by a motion capture system, leading to significantly improved flexibility and accuracy, as well as reduced workload and complexity in lower limb movement evaluation. Following laboratory validation, the calibrated sensor network is deployed in clinical settings for assessing lower limb function in patients with hemiplegia. The proposed calibration method and sensor network are empirically evaluated to provide a robust theoretical and technical foundation for clinical application. By establishing a reliable posture evaluation network tailored for rehabilitation training and potential for home monitoring, this study offers innovative calibration strategies and practical solutions that support the optimization of rehabilitation assessment and treatment, contributing to the advancement of intelligent rehabilitation medicine.