Presentation on theme: "ISSUES IN 3D OBJECT RECOGNITION Jean Ponce Department of Computer Science and Beckman Institute University of Illinois at Urbana-Champaign Joint work with."— Presentation transcript:
ISSUES IN 3D OBJECT RECOGNITION Jean Ponce Department of Computer Science and Beckman Institute University of Illinois at Urbana-Champaign Joint work with Amit Sethi, David Renaudie and David Kriegman and Svetlana Lazebnik, Cordelia Schmid and Martial Hebert
Face Camel Bug Human/Felix Joe Barbara Steele Problem: Recognizing instances Recognizing categories
Variability : Camera position Illumination Internal parameters Within-class variations
Question #1: Is it better to eliminate as many possible of the parameters that govern appearance or is it better to work with the raw pixels? Note: We may know something about the “shape” and the “dimension” of our image set. This “surface” is not smooth.
Brooks and Binford, 1981 Sullivan and Ponce, 1998 Murase and Nayar, 1992 Schmid and Mohr, 1996 Invariants (Weiss, 1988; Rothwell et al., 1992; etc.)
Face Camel Bug Human ??
Question #2: What is an appropriate object representation for describing people, animals, chairs, boats, shoes, etc. ?? Do we really believe that local pixel signatures and their geometric/statistical relationships are sufficient? or
The Blum transform, 1967 Generalized cylinders Binford, 1971
Zhu and Yuille, 1996
Question #3: How do we construct object descriptions from images? = How do we segment images? = How do we compute our feature vectors?
Question #4: How can we formalize the object recognition process? What should the corresponding optimization process try to optimize? or
d1d2 d3 d1 d2 d3 The pedal curve The trace
d1 d2 d3 d
d3 Occluding contour Silhouette
Question #5: How can we effectively deal with clutter ?
Baseline Frontier point 3D/4D 1 3 (Cipolla, Åström and Giblin, 1995)
How do we recognize objects at the category level? Question #6: