Recognizing human motion using parameterized models of optical flow

1997

Book Chapter

ps

The tracking and recognition of human motion is a challenging problem with diverse applications in virtual reality, medicine, teleoperations, animation, and human-computer interaction to name a few. The study of human motion has a long history with the use of images for analyzing animate motion beginning with the improvements in photography and the development of motion-pictures in the late nineteenth century. Scientists and artists such as Marey [12] and Muybridge [26] were early explorers of human and animal motion in images and image sequences. Today, commercial motion-capture systems can be used to accurately record the 3D movements of an instrumented person, but the motion analysis and motion recognition of an arbitrary person in a video sequence remains an unsolved problem. In this chapter we describe the representation and recognition of human motion using parameterized models of optical flow. A person's limbs, face, and facial features are represented as patches whose motion in an image sequence can be modeled by low-order polynomials. A robust optical flow estimation technique is used to recover the motion of these patches and the recovered motion parameters provide a rich, yet concise, description of the human motion which can be used to recognize human activities, gestures, and facial expressions.

@incollection{black:MBR:1997,
  title = {Recognizing human motion using parameterized models of optical flow},
  author = {Black, M. J. and Yacoob, Y. and Ju, X. S.},
  booktitle = {Motion-Based Recognition},
  pages = {245-269},
  editors = {Mubarak Shah and Ramesh Jain,},
  publisher = {Kluwer Academic Publishers},
  address = {Boston, MA},
  year = {1997},
  doi = {}
}