A Constrained-Based Optimization Method for Real-Time Kinematics Using Magneto-Inertial Signals: Application to Upper Limb Joint Angles Estimation During Prolonged Recordings

This work presents a flexible method for the real-time estimation of human joint angles from magneto-inertial measurement technology. The method aims to enhance the accuracy and consistency of joint angle estimates by incorporating physiological joint limits and task-specific motor characteristics into the optimization process, thanks to a biomechanical model. As an explanatory example, the method was applied to shoulder and elbow joints during a prolonged writing task. The adopted upper limb model was designed following the International Society of Biomechanics guidelines and the Denavit-Hartenberg convention, ensuring anatomical relevance and computational efficiency. By comparing results with stereophotogrammetric tracking outputs, the application of constraints - leveraging a priori knowledge of the workspace boundaries for joint centers - enhanced the accuracy of shoulder and elbow angle estimations and effectively mitigated the impact of sensor orientation drift over extended periods. This method ensured that joint centers trajectories remain within task-specific workspace limits, thus preventing deviations that are not compatible with the expected kinematic behavior. The percentage decrease in the root mean square average errors amounted to about 13% in the time intervals when constraints were active, demonstrating the method's effectiveness in reducing the errors. Computationally time-wise, joint angles were estimated with an update period of about 10 ms, allowing real-time usage. The proposed method can be easily generalized to different biomechanical models and to include information from complementary technologies, making it applicable across various contexts such as clinical assessments, rehabilitation, and ergonomics.