🚀 go-pugleaf

                     *** $B9V1i2q$N$*$7$i$;(B ***

  $B?7>pJs=hM}3+H/5!9=$G$O%3%s%T%e!<%?%S%8%g%s$K4XO"$7$?9V1i2q$r9T$$(B
$B$^$9!#D09V<+M3$G$9$N$G!"$46=L#$r$*;}$A$NJ}$O$<$R$*$$$G$/$@$5$$!#(B

  $B1i<T$O8=:_(B ATR $B$KBZ:_$5$l$F$$$k(B INRIA $B$N(B Zhengyou Zhang $B@h@8$G$9!#(B

$BF|;~(B: 1997 $BG/(B 11 $B7n(B 6 $BF|(B ($BLZ(B)
      14:30 - 16:30        Lecture by Prof. Zhengyou Zhang 

$B>l=j(B: RWCP $B$D$/$P8&5f%;%s%?!<(B 13F $B2q5D<<(B
      ($B0q>k8)$D$/$P;T(B $B$D$/$P;00f%S%kFb(B)
      http://www.rwcp.or.jp/view/general/rwc-map/home.html

Title: Computer Vision, Photorealistic Modeling, and Visual Learning

Abstract:

In this talk, I will give an overview of my work on image matching,
motion analysis, 3D modeling and visual learning. Many examples from
real applications will be used to demonstrate the applicability of the
developed techniques. Emphasis will be on the use of projective geometry
with uncalibrated images. Use of uncalibrated images provides us with a
lot of flexibilities (camera parameters can be changed either
accidentally or volunteerly; images can be taken with different cameras;
etc.). Furthermore, the recovered projective or affine structure
contains information rich enough for performing certain tasks such as
navigation and object recognition, and can be brought back to the
Euclidean space if a little bit of domain knowledge is injected.

The following topics will be addressed in more details:
- multiple-view geometry, weak- and self-calibration
  (a unified approach to dealing with perspective and affine cameras)
- building a CAD-like representation of a scene/object
- synthesizing new views without 3D models
- Recovering Euclidean shape from uncalibrated images 
  with fuzzy domain knowledge
- modeling illumination variation from images

If time allows, ongoing work on visual learning (with the application to
facial expression recognition) will also be presented. 



-- 
$B9b66M5?.(B($B$?$+$O$7$R$m$N$V(B): $B?7>pJs=hM}3+H/5!9=(B $B$D$/$P8&5f%;%s%?!<(B
E-mail: hironobu@trc.rwcp.or.jp or hironobu@tiny.or.jp