doi:10.1016/S0031-3203(02)00126-7 
Copyright © 2002 Pattern Recognition Societyc. Published by Elsevier Science B.V.
Pedestrian registration in static images with unconstrained background
Department of Computer Science, School of Computing, National University of Singapore, Singapore 119260, Singapore
Received 26 November 2001;
revised 21 May 2002.
This article is dedicated in memory of Kah-Kay Sung
Available online 12 December 2002.
References and further reading may be available for this article. To view references and further reading you must
purchase this article.
Abstract
This paper introduces a human body contour registration method for static pedestrian images with unconstrained backgrounds. By using a statistical compound model to impose structural and textural constraints on valid pedestrian appearances, the matching process is robust to image clutter. Experimental results show that the proposed method register pedestrian contours in complex backgrounds effectively.
Author Keywords: Human body registration; Statistical modeling; Similarity measure; Feature extraction; Image matching
Fig. 1. Initial pedestrian training images. Notice the differenceS in clothing texture, varying poses and different lighting conditions. Also notice the different image backgrounds (e.g. zebra crossing in (c) and (f), trees in (a), (b) and (d), other pedestrians in (c) and (f), shadows in (e) and (f)).
Fig. 2. A model-based pedestrian contour registration approach consists of three components: (1) a compound pedestrian model
IM(α,γ) capturing permissible image variations; (2) a combined feature-texture similarity measure accounting for image differences between the model image and given pedestrian images; and (3) an iterative pedestrian registration algorithm to find the best model parameters corresponding to the minima of the proposed similarity measure.
Fig. 3. Manual registration of training images: (a) Prototype body contour (white dots represent head, hands and feet); (b) Sparse contour point model; (c) Training image manual registration; and (d) Feature point correspondence map (FPCM) between (c) and (b).
Fig. 4. Preprocessed pedestrian training images which have structural variation removed. Also, the image backgrounds are masked.
Fig. 5. Pedestrian textural variation eigenvectors: (a) the mean pedestrian image; (b)–(g) Eigenvectors 1–6.
Fig. 7. Pedestrian structural variation eigenvectors. Left–right: eigenvectors 1–6.
Fig. 9. Synthesized pedestrian shapes.
Fig. 10. Nonlinear distribution of varying pose face images. Each data represents a synthesized face image projected in the subspace spanned by the 3 most significant eigenvectors. pose 1: right rotated; pose 2: slightly right rotated; post 3: frontal; post 4: slightly left rotated; and pose 5: left rotated.
Fig. 6. Graphs of eigenvalues for example structurally normalized pedestrian images. (
Nλ=20) is selected for this plot.
X-axis: index of sorted eigenvalues;
Y-axis: eigenvalues. Note that the majority of variation is captured in the first
Nλ eigenvectors.
Fig. 8. Graphs of eigenvalues for example FPCMs of pedestrian images. (
Nμ=15) is selected for this plot.
X-axis: index of sorted eigenvalues;
Y-axis: eigenvalues. Note that the majority of variation is captured in the first
Nμ eigenvectors.
Fig. 11. Pedestrian registration results. Row 1 and 2: good alignments. Row 3: fair alignments. Note that severe clutter (e.g. zebra crossing, trees, and other pedestrians, etc.) are present in the backgrounds.
Fig. 12. Pedestrian mis-registration examples. Image (a) is misregistered because it is too blurry for reliable features to be extracted. The people in image (b) is misregistered because his left hand is placed on his head, exhibiting too much pose variation for the learnt pedestrian model.
Fig. 13. Human body registration results with complex scene.
Table 1. Pedestrian registration results, see the goodness criterion defined in Section 4.1
Abstract
Purchase PDF (1012 K)
E-mail Article
Add to my Quick Links
Cited By in Scopus (2)
Purchase PDF (1178 K)
