Loading...

Vision-based 3d Object Recognition

Forghani, Hossein | 2014

522 Viewed
  1. Type of Document: M.Sc. Thesis
  2. Language: Farsi
  3. Document No: 46201 (19)
  4. University: Sharif University of Technology
  5. Department: Computer Engineering
  6. Advisor(s): Kasaei, Shohreh
  7. Abstract:
  8. Object recognition is a field of image processing with the purpose of object detection and determining its pose. There was a wide research of object recognition using 2d information. These methods are less efficient due to their sensitivity to view and scale. Today with development of 3d technology, 3d reconstruction of objects has been eased, thus there is a new trend of 3d methods. 3d object recognition uses 3d data and model to recognize objects more efficiently in different views and in occlusion and clutter. The related works are classified in three category of view-based, feature matching, and classification. In this research we attempted to improve one of the best feature matching methods. This method first identifies correspondences between 2d image features and 3d points of model. Then filters the correspondences in two local and gloabl stages. In the local filtering only the neighboring correspondences are considered and finally a few best hypotheses are returned. In global filtering the consistency of all correspondences are checked. Finally in the pose estimation stage, RANSAC is guided by correspondences compatibility graph to estimate the object 3d poses. We leveraged graph matching to improve the method. Graph matching is the problem of matching members of two feature sets. In the proposed method, filtering is done by two stages of graph matching using local and global criteria. In the global filtering, a new angle consistency criterion is introduced. A new method of graph matching based on grdient ascent is developed that gets a better result than current graph matching methods. The experimental results on Microsoft Research dataset shows our proposed method increases the recall by 0.02 while the precision is almost 1 in the cost of 40% growth in the computational time on average
  9. Keywords:
  10. Descriptor ; Classification ; Correspondence Theory ; Three Dimentional Objects Recognition ; Feature Matching

 Digital Object List

 Bookmark