3D dynamic pose estimation from marker-based optical data

Abstract

The desire to capture images of human movement has existed since prehistoric times (see chapter "Observing and Revealing the Hidden Structure of the Human Form in Motion Throughout the Centuries"). However, it is only since the late nineteenth century and the development of cameras able to capture multiple sequential images that the recording and quantitative analysis of movement has become possible. With modern cameras and high computational power now available, it is commonplace for researchers and clinicians to make detailed measurements, from which an estimation of the position and orientation (pose) of a human body during motion can be computed. This chapter focuses on the estimation of dynamic 3D pose based on optical motion capture systems that record the 3D location of markers attached to the body (see Fig. 1). In this chapter, we describe the estimation of the pose of a multibody model comprising segments that are connected by joints that constrain the direction and range of motion between those segments. There are three common deterministic solutions to the problem of pose estimation; direct, single body, and multibody. This chapter focuses on the two optimization methods, single body and multibody, that provide a deterministic and a discriminative solution to the problem of pose estimation. Unlike the direct pose estimation, these two approaches mitigate, to some extent, uncertainty in the data.