Premium
Object detection using straight line matching in θ‐ρ space
Author(s) -
Okuzono Taisei,
Wakizako Hitoshi
Publication year - 2010
Publication title -
electronics and communications in japan
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.131
H-Index - 13
eISSN - 1942-9541
pISSN - 1942-9533
DOI - 10.1002/ecj.10176
Subject(s) - hough transform , artificial intelligence , matching (statistics) , computer vision , scale invariant feature transform , invariant (physics) , computation , rotation (mathematics) , line (geometry) , computer science , mathematics , pattern recognition (psychology) , edge detection , algorithm , geometry , feature extraction , image processing , image (mathematics) , statistics , mathematical physics
The contours of many industrial parts contain straight lines and the positions of the lines are therefore useful information for object detection. This paper presents a matching technique for straight lines. The method consists of θ‐matching, ρ‐matching, and pose estimation. Any lines in 2D space are represented with parameters θ and ρ by the Hough transform. In order to find the corresponding lines in a model and a scene, the θ and ρ values are evaluated in θ‐matching and ρ‐matching. When an object is translated and rotated, the contour lines of the object are also transferred and the θ values of the lines are merely shifted by the rotation angle in the θ‐ρ space. Thus, the relative positions of the θ values are invariant. In θ‐matching, the corresponding lines of the model and the scene are selected so that the relative θ values of the corresponding lines are nearly equal. In ρ‐matching, the corresponding lines are evaluated further by computing the deviations of their ρ values. Finally, the transfer parameters of the selected pairs are estimated in pose estimation. The experiments show that this technique is robust to rotation, occlusion, and scaling of the objects. We also discuss the computation time, in which the preprocess such as edge detection and the Hough transform takes much of the time. © 2010 Wiley Periodicals, Inc. Electron Comm Jpn, 93(3): 34–41, 2010; Published online in Wiley InterScience ( www.interscience.wiley.com ). DOI 10.1002/ecj.10176