Modeling People Toward Vision-Based Understanding of a Person's Shape, Appearance, and Movement

Automated capture of human shape, motion, and behavior has been a long-standing goal of computer vision research. Research has been driven by numerous potential applications of such a technology in the areas of man–machine interfaces, visual surveillance, biometrics, and media content production. There currently exist partial solutions, such as optical motion capture or laser-based shape capture. However, they tend to be both expensive and cumbersome. Imageor video-based capture offers the promise of a cheaper and more flexible alternative. Over the past decade, significant progress has been made toward the goal of reliably and automatically interpreting videos of people. However, because images are inherently noisy and cluttered, significant obstacles remain. This special issue brings together eight papers that illustrate the current state of the art, present the most recent advances in the field, and introduce the remaining open issues. Moeslund and Granum survey the various developments in vision-based human modeling over the past two decades. This paper provides a comprehensive taxonomy of the field that complements earlier reviews [1, 3]. The authors discuss the stages common to all vision-based techniques: Initialization, tracking, pose estimation, and recognition. They highlight the progress made toward the development of systems for noninvasive capture of human movement and importantly identify common assumptions and limitations of existing approaches.