Model-Based Human Motion Capture

Abstract

Human motion analysis is a challenging research area aimed at automating the study of human behavior. An important part of any such system is the component that performs the Human Motion Capture (HMC); in order for human motion to be processed and semantically analyzed, a mathematical representation of the observed motion needs to be extracted. There are two separate aspects to a HMC system; sensing (hardware) and processing (software). The processing itself comprises of an initialization (anthropometry and pose estimation) and a tracking phase. In this chapter, we present methods for three-dimensional model-based human motion capture from uncalibrated passive optical sensors with semi-antomatic initialization and tracking. Such methods allow for non-untrusive capture of natural human behavior from video cameras or from archival recordings. We demonstrate the accuracy, advantages, and limitations of our methods for various classes of data.