A Novel Cascaded Multimodal Information Fusion Approach for Stable Real-Time Lower Limb Motion Recognition via Portable Sensors

Accurate gait phase estimation and recognition of lower limb movements are crucial for wearable assistive devices in medical rehabilitation. Due to the complexity of movement scenarios, obtaining precise motion modes and gait phases in real-time is particularly challenging. This is especially true for tasks that require high responsiveness, where algorithms need to recognize gait phase and movement mode simultaneously with or even ahead of the motion to inform further decisions by wearable devices. To address this issue, we proposed a novel cascaded multimodal information fusion approach for stable real-time lower limbmotion recognition based on a lightweight lower limb sensor system integrating shank-mounted inertial measurement unit (IMUs) and foot pressure sensors. Furthermore, an adaptive extended Kalman filter (AKEF) was employed to enhance the accuracy of movement mode recognition. In experimental settings, the continuous gait phase estimation error in mixed movement scenarios was as low as 2.12%, with errors of 1.43% in the stable gait phase and 2.51% in the transition phase. The movement mode recognition accuracy in mixed movement scenarios improved from 98.42% to 99.21%. This study will be applied to subsequent research in human motion analysis and real-time control of lower limb exoskeletons.