Robust object recognition using local appearance based methods
In this thesis we present an approach to appearance-based object recognition using single camera images. The approach is based on using an attention mechanism to obtain visual features that are generic, robust and informative. The features themselves are recognized using principal components in the...
| Main Author: | |
|---|---|
| Other Authors: | |
| Format: | Others |
| Language: | en |
| Published: |
McGill University
2000
|
| Subjects: | |
| Online Access: | http://digitool.Library.McGill.CA:80/R/?func=dbin-jump-full&object_id=33472 |
| Summary: | In this thesis we present an approach to appearance-based object recognition using single camera images. The approach is based on using an attention mechanism to obtain visual features that are generic, robust and informative. The features themselves are recognized using principal components in the frequency domain. We show that we can perform robust appearance based object recognition by using the visual characteristics of only a small number of such features. The technique is robust to planar translations and rotations of the object being recognized due to our polar sampling in the frequency domain. We are able to recognize objects on different types of background clutter due to a masking technique we've developed. We are also able to handle a degree of occlusion as we make use of multiple features for the purposes of recognition. === The same approach is further applied in the field of robotics to provide a means for the automatic recognition of locations or landmarks in scenes typically encountered by mobile robots. Hence instead of only recognizing objects, we also present a means of using the same computational model to recognize locations, thus performing coarse localization. |
|---|