THE EVOLUTION OF POSE ESTIMATION ALGORITHMS FOR 3D MOTION CAPTURE DATA: COPING WITH UNCERTAINTY

Abstract

At the heart of many biomechanical analyses is the estimation of the pose (position and orientation) of a multi-segment model based on recording of 3D motion data. The principle assumption of most pose estimation algorithms is that sensors move rigidly with the body segments to which they are attached. It is accepted, however r, that sensors attached to the skin move e relative to the underlying skeleton and that this idiosyncratic Soft Tissue Artifact (STA A) is challenging to model. Usually pose is estimated with discriminative algorithms that are ill-suited to the uncertainty of STA. Emerging algorithms based on probabilistic inference may mitigate STA by encoding the pose e and any prior knowledge about the pose e probabilistically, and capture the “artifacts” using a generative model.