Research on Quaternion Pose Feature Extraction and Its Application in Intention Recognition

In order to circumvent the problem of deadlock and discontinuity inherent in Euler angles, this study aims to explore pattern recognition characteristics based on quaternions for accurately discerning the motion intentions of wearers of exoskeleton robots in specialized environments. We devised an experimental framework based on common underwater motion patterns, comprising signal acquisition, signal preprocessing, feature selection, and extraction procedures. In the process of feature extraction, a kind of angle feature containing joint position relation is found, and its mathematical formula expressed in quaternion is derived. Finally, through comparative pattern recognition experiments employing two machine learning algorithms on both original data features and joint angle features, the feasibility of employing quaternions as pattern recognition features was validated.