Efficient Prediction of Segment Kinematics and Dynamics from Motion Capture Data Using Deep Learning

In recent years, motion capture systems have emerged, allowing for a broad and ever-expanding range of practical applications in a myriad of domains, including healthcare and biomechanics, sports, and artificial reality technologies. These concise and deterministic systems, however, suffer from limitations of prohibitive costs, and the practical infeasibilities of being used en-masse for round-the-clock day-to-day applications. This paper investigates the possibility of drastically reducing the number of sensors required to satisfactorily predict, within acceptable bounds of error, a range of targeted kinematic and dynamic particulars endemic to a specific domain of work, and determining the optimal placements of these sensors on the human body. These are achieved through a judicious application of data-analysis techniques, followed by the training and testing of artificial neural networks to establish a foundational proof of concept that may serve as a basis for further research.