doi:10.1016/S0031-3203(02)00046-8 
Copyright © 2002 Pattern Recognition Society. Published by Elsevier B.V.
Fast object recognition using dynamic programming from combination of salient line groups
a Department of Robot System Engineering, Tongmyong University of Information Technology, Yongdang-dong 535, Nam-gu, Busan City, South Korea
b Samsung Corning Industry Corporation, Mechatronics Group, Kweonsun-gu, Suwon City, South Korea
c Department of Electrical Engineering, Korea Advanced Institute of Science and Technology, 373-1, Gusong-dong, Yusong-gu, Daejun, South Korea
Received 21 August 2001;
accepted 28 January 2002.
Available online 17 February 2006.
References and further reading may be available for this article. To view references and further reading you must
purchase this article.
Abstract
This paper presents a new method of grouping and matching line segments to recognize objects. We propose a dynamic programming-based formulation extracting salient line patterns by defining a robust and stable geometric representation that is based on perceptual organizations. As the endpoint proximity, we detect several junctions from image lines. We then search for junction groups by using the collinear constraint between the junctions. Junction groups similar to the model are searched in the scene, based on a local comparison. A DP-based search algorithm reduces the time complexity for the search of the model lines in the scene. The system is able to find reasonable line groups in a short time.
Author Keywords: Feature matching; Dynamic programming; Perceptual grouping; Object recognition
Fig. 2. Binary relations made from any two connected junction nodes: (a) line segments on a model; and (b) combination of junctions by perceptual constraints, such as proximity, collinearity, and parallelism.
Fig. 3. The DP algorithm searches a scene node corresponding to each model node. A model feature can be matched to at least one node, among scene nodes, 1,…,m+1 of a column including NULL node (NL). (a) Line segments for the rear view of a vehicle; and (b) A DP-based search. m is the number of junctions detected from (a) and M is the number of pre-defined model junctions.
Fig. 4. Junction extractions: the number of junctions depends on the condition of images. Each column consists of an original image, line segments, and junctions and their intersecting points for quality measure 0.5, respectively. The small circle in the figure presents the intersection point of two-line junction. (a) parts: 2-D scene under a controlled lighting condition; (b) blocks: an indoor image with good lighting; and (c) cars: a cluttered image under uncontrolled outdoor road scene.
Fig. 5. Occupying percentage of junctions according to the change of the quality measure.
Fig. 6. An object matching under weak perspective projection: a rear-window of a vehicle on the highway is used: (a-1) the first and last images to be tested; (a-2) line extraction; (a-3) junction detection for QJ=0.5; (a-4) optimal model matching; and (b) a few optimal matching results between the first and last images.
Fig. 7. Object matching in a synthetic image with broken and noisy lines.
Fig. 8. Topological shapes for model description: (a) a model with clockwise rotation in the starting junction; and (b) a model with counterclockwise direction in the starting junction.
Fig. 9. A topological shape extraction for 3-D object recognition. (a) original image; (b) line extraction; and (c) found topological shapes.
Fig. 10. Collinear lines according to the change of standard deviation σ0 as a threshold. (a) randomly generated original line set; Line pairs detected for; (b) σ0=0.4; (c) σ0=0.3; and (d) σ0=0.1.
Table 1. Junction number vs. quality measure
Table 2. The topological matching results
Abstract
Purchase PDF (453 K)
E-mail Article
Add to my Quick Links
Cited By in Scopus (3)
Purchase PDF (2275 K)
