The Design and Use of Steerable Filters
IEEE Transactions on Pattern Analysis and Machine Intelligence
Tangible bits: towards seamless interfaces between people, bits and atoms
Proceedings of the ACM SIGCHI Conference on Human factors in computing systems
Feature Detection with Automatic Scale Selection
International Journal of Computer Vision
Object Recognition Using Coloured Receptive Fields
ECCV '00 Proceedings of the 6th European Conference on Computer Vision-Part I
Local Scale Selection for Gaussian Based Description Techniques
ECCV '00 Proceedings of the 6th European Conference on Computer Vision-Part I
Hi-index | 0.00 |
Physical icons (phicons) are ordinary objects that can serve as user interface in an intelligent environment. This article addresses the problem of recognizing the position and orientation of such objects. Such recognition enables free manipulation of phicons in 3D space. Local appearance techniques have recently been demonstrated for recognition and tracking of objects. Such techniques are robust to occlusions, scale and orientation changes. This paper describes results using a local appearance based approach to recognize the identity and pose of ordinary desk top objects. Among the original contributions is the use of coloured receptive fields to describe local object appearance. The view sphere of each object is sampled and used for training. An observed image is matched to one or several images of the same object of the view sphere. Among the difficult challenges are the fact that many of the neighborhoods have similar appearances over a range of view-points. The local neighborhoods whose appearance is unique to a viewpoint can be determined from the similarity of adjacent images. Such points can be identified from similarity maps. Similarity maps provide a means to decide which points must be tested to confirm a hypothesis for correspondence matching. These maps enable the implementation of an efficient prediction-verification algorithm. The impact of the similarity maps is demonstrated by comparing the results of the prediction-verification algorithm to the results of a voting algorithm. The ability of the algorithm to recognize the identity and pose of ordinary desk-top objects is experimentally evaluated.