Recognizing planar objects using affine line invariants by geometric hashing = 선형 직선 불변량과 기하학적 해싱에 의한 평면 물체의 인식

Recognition of industrial parts and their locations in a factory environment is a major task in robot vision. We use an efficient recognition algorithm, called Geometric Hashing, which assumes the affine approximation to the perspective viewing transformation. This technique is based on two stages: the first stage is an intensive model preprocessing stage, done off-line, where transformation invariant features of the models are indexed into a hash table. The second is an actual recognition stage, which employs the efficient indexing made by the above technique. As the number of models in a model database, however, becomes larger, the size of search space to find a corresponding model may increase exponentially. In order to solve this problem, we introduce a line convex hull (LCH). The line convex hull classifies a set of four lines into one of six different types of convex hulls and also assigns possible orderings to the four lines. By using these types of line convex hulls, It has been found that distinguishing between different types of a set of four lines in a model or scene results in an efficient implementation of Geometric Hashing using multidimensional hash tables. The algorithm was tested in recognition of industrial objects of welding panels. We have implemented the recognition system described and carried out a series of experiments on real images. By combining the LCH with the Geometric Hashing technique, the algorithm greatly reduced the search space to find a candidate model instance. The system has successfully recognized and localized the models in cluttered scenes from the database composed of ten models.
Kweon, In-Soresearcher권인소researcher
Issue Date
113255/325007 / 000957029

학위논문(석사) - 한국과학기술원, 자동화및설계공학과, 1997.2, [ v, 69 p. ]


인식; 불변량; Geometric hashing; 기하학적 해싱; Recognition; Invariant

Appears in Collection
Files in This Item
There are no files associated with this item.
  • Hit : 100
  • Download : 0
  • Cited 0 times in thomson ci


  • mendeley


rss_1.0 rss_2.0 atom_1.0