Current research suggests the emergent need to recognize and predict locomotion modes (LMs) and LM transitions to allow a natural and smooth response of lower limb active assistive devices such as prostheses and orthosis for daily life locomotion assistance. This Master dissertation proposes an automatic and user-independent recognition and prediction tool based on machine learning methods. Further, it seeks to determine the gait measures that yielded the best performance in recognizing and predicting several human daily performed LMs and respective LM transitions. The machine learning framework was established using a Gaussian support vector machine (SVM) and discriminative features estimated from three wearable sensors, namely, inertial, force and laser sensors. In addition, a neuro-biomechanical model was used to compute joint angles and muscle activations that were fused with the sensor-based features. Results showed that combining biomechanical features from the Xsens with environment-aware features from the laser sensor resulted in the best recognition and prediction of LM (MCC = 0.99 and MCC = 0.95) and LM transitions (MCC = 0.96 and MCC = 0.98). Moreover, the predicted LM transitions were determined with high prediction time since their detection happened one or more steps before the LM transition occurrence. The developed framework has potential to improve the assistance delivered by locomotion assistive devices to achieve a more natural and smooth motion assistance.
展开▼
