doi:10.1016/S0262-8856(02)00006-9 
Copyright © 2002 Elsevier Science B.V. All rights reserved.
How to measure the pose robustness of object views
Gabriele Peters
, 
, a, Barbara Zitovab and Christoph von der Malsburga, 1
a Institut für Neuroinformatik, Systembiophysik, Ruhr-Universität Bochum, Universitätsstr. 150, D-44780 Bochum, Germany
b Department of Image Processing, Institute of Information Theory and Automation, Academy of Sciences of the Czech Republic, Pod vodárenskou v
í 4, 182 08 Praha 8, Czech Republic
Received 16 October 2000;
accepted 18 December 2001.
Available online 8 February 2002.
References and further reading may be available for this article. To view references and further reading you must
purchase this article.
Abstract
The viewing hemisphere of a three-dimensional object can be partitioned into areas of similar views, which provide pose robustness. We compare two procedures for measuring the robustness of views to pose variation: tracking of object features, i.e. Gabor wavelet responses, by utilizing the continuity of successive views and matching of features in different views, which are assumed to be independent. Both procedures proved to be appropriate to detect canonical views. We found no difference concerning the size of the view bubbles, but tracking provides more precise correspondences than matching. Tracking is more appropriate for recognizing changes of features, whereas matching is more suitable if features of the same appearance are to be found.
Author Keywords: Three-dimensional object perception; Pose robustness; Matching/tracking object features; Canonical views
Fig. 1. Viewing hemisphere with examples of view bubbles. The representation of a viewing hemisphere consists of 100×25 views. Each crossing of the grid stands for one view. The angle between two neighboring views is 3.6° in either direction. The dot in front marks view (0,0). The depicted view bubbles have been determined with the tracking procedure for object Tom. View (79,14) provides a small view bubble. It includes views which cover a range of 21.6° in x-direction and 14.4° in y-direction. View (6,6) provides a larger view bubble, which covers a range of 43.2 and 28.8°, respectively.
Fig. 2. Preprocessing: (a) original image, (b) gray level segmentation, (c) centered segmentation after eliminating wrong segments, (d) original image masked with the result of centered segmentation, (e) grid graph on object segment, and (f) grid graph on original image.
Fig. 3. Distribution of view similarities for object Tom.
Fig. 4. Distribution of view similarities for object Dwarf.
Fig. 5. Correspondences for object Tom: solid line represents tracked sequence, and dashed line represents matched sequence. (a) First poor match.
Fig. 6. Correspondences for object Dwarf: solid line represents tracked sequence, and dashed line represents matched sequence. (a) First poor match.
Fig. 7. Qualitative similarity diagram: ‘good’ and ‘poor’ is meant in the sense of correct, respectively, incorrect, correspondences. See description in the text for details.
Fig. 8. Canonical and non-canonical views for object Tom. View (47,6) is the view with the largest area of its view bubble (generated by tracking). Compare with the first diagram of Fig. 3.
Abstract
Purchase PDF (1533 K)
E-mail Article
Add to my Quick Links
Cited By in Scopus (1)
Purchase PDF (487 K)
