Ambulatory inertial spinal tracking using constraints

Wearable inertial sensor networks represent a well-known and meanwhile cheap solution for in-field motion capturing. However, the majority of existing approaches and products rely on simple stick figure models to approximate the human skeleton with only a few rigid segments and connecting joints. Especially the spine is often extremely simplified with one or at most two segments. This simplification results in significant kinematic estimation errors. This paper presents a novel inertial tracking approach, where a recursive filter with integrated constraints enables detailed and efficient estimation of the spine kinematics in real time. The advantages of the proposed approach are confirmed in experiments using ground truth data from an optical reference system.