Grouping Iso-Velocity Points for Ego-Motion Recovery

Abstract

The instantaneous image motion field due to a camera moving through a static environment encodes information about ego-motion and environmental layout. For pure translational motion, the motion field has a unique point termed focus of expansion/contraction where. the image velocity vanishes. We reveal the fact that for an arbitrary 3D motion the zero-velocity points, whose number can be large, have the regularity of being approximately cocircular. More generally, all the image points with the same velocity u are located approximately on a circle (termed the isovelocity circle (IVC)) determined solely by u and the ego-motion, except for the pathological cases in which the circle degenerates into a straight line. While IVCs can be recovered from 3 or more pairs of iso-velocity points in the motion field using a linear method, estimating ego-motion reduces. to solving systems of linear equations constraining iso-velocity point pairs (Yang 1992).

Cite

Text

Yang and Yuille. "Grouping Iso-Velocity Points for Ego-Motion Recovery." AAAI Conference on Artificial Intelligence, 1992.

Markdown

[Yang and Yuille. "Grouping Iso-Velocity Points for Ego-Motion Recovery." AAAI Conference on Artificial Intelligence, 1992.](https://mlanthology.org/aaai/1992/yang1992aaai-grouping/)

BibTeX

@inproceedings{yang1992aaai-grouping,
  title     = {{Grouping Iso-Velocity Points for Ego-Motion Recovery}},
  author    = {Yang, Yibing and Yuille, Alan L.},
  booktitle = {AAAI Conference on Artificial Intelligence},
  year      = {1992},
  pages     = {356-361},
  url       = {https://mlanthology.org/aaai/1992/yang1992aaai-grouping/}
}