Nevada - 9
Figure 8(a) Figure 8(b)
Figure 8. Projection of the 3D model of the scene in Figure 1(a) and
the model registered with an image
descriptions. Lower level descriptions, such as line segments can give rise to a very large number
of possible matches. It is important that we be able to segment the features corresponding to one
object from those that come from other objects or background. Many model-based methods do
exist that work only by matching lower level features [5], however, such systems have high
computational complexity and can work only with a small number of models. Techniques using
higher level, surface or object level descriptions [10] can handle more objects and offer much
more efficient recognition. However, these approaches are dependent on robust construction of
the higher level descriptions. Work on such approaches is still in early stages of progress.
5 Acknowledgments
Andres Huertas, Chungan Lin, Sanjay Noronha and Mourad Zerroug, my colleagues, have
helped with the preparation of this paper and provided examples used herein. I wish to
acknowledge support of the Advanced Research Projects Agency (ARPA) of the Department of
Defence under contract numbers F49620-93-1-0620 and DACA76-93-C-0014 monitored by the
Air Force Office of Scientific Research and by the Topographic Engineering Center respectively.
6 Bibliography
[1] M. Bejanin, A. Huertas, G. Medioni, R. Nevada, “Model Validation for Change Detection”,
WACV-94, pp. 160-167.
[2] R.C.K. Chung and R. Nevada, “Recovering building structures from stereo”, ACV-93, pp. 64-
73.
