Human Motion Intention Recognition with Thigh Mounted IMUs

Assistive devices such as powered prosthesis have the potential to help people with amputations by restoring mobility and contributing to their independence. The ability to recognize human intentions and preceding actions is critical to the smooth and safe interaction between a user and the powered prosthesis. Although many methods have been proposed to recognize the user's intended motion, most methods rely on multiple sensors placed on various body segments, making them impractical for many real-world applications. In addition, sliding-window-based methods are typically employed for analysis and feature extraction. However, the inability to adapt the window size to varying walking speeds limits the performance of most existing methods in real applications. To address these challenges, this study proposes a novel intention recognition approach based solely on thigh-mounted inertial measurement units (IMUs) and an event-based feature extraction approach that achieved a good generalization performance across different speeds and subjects. The proposed method was evaluated on 10200 steps of level ground walking, stairs ascent, and stairs descent activities from 10 healthy subjects. This system achieves an accuracy greater than 99% for all test subjects in the Leave-One-Subject-Out validation. In addition, activities were recognized between 150.63 to 724.32 milliseconds prior to the next foot contact event. These results demonstrate the potential of the proposed method in recognizing the motion intention of a user to enable optimized control of the powered prosthesis in time.