Performance Evaluation of Vision-Based High DOF Human Movement Tracking: A Survey And Human Computer Interaction Perspective

Abstract

Accurate tracking of human movement has the potential to enable many kinds of human-computer interaction (HCI). Vision-based methods promise tracking without encumberance by body-mounted apparatus but constitute a significant research challenge. Its many applications both in HCI and other areas such as surveillance have made markerless human tracking a central problem in computer vision. Despite this research focus, there are few established mechanisms for evaluating and comparing the performance of reported solutions. Furthermore, very few systems have attempted human-centered evaluation using methods from HCI research. This paper presents a review of evaluation methods employed in the literature of high degree of freedom human movement tracking systems. Detailed full-body tracking and hand tracking systems are included. Suitable evaluation methods employed in other computer vision and HCI research are also considered.