Object Tracking: Feature Selection and Confidence Propagation

Authors:
Affiliations:
Venue:
CRV '04 Proceedings of the 1st Canadian Conference on Computer and Robot Vision
Year:
2004

Citing 2
Cited 1

Digital video processing

Digital video processing
Kernel-Based Object Tracking

IEEE Transactions on Pattern Analysis and Machine Intelligence

Unscented feature tracking

Computer Vision and Image Understanding

Quantified Score

Hi-index	0.00

Visualization

Abstract

Choosing unique and invariant features is the firstimportant step in object tracking. In this paper, wepresent a method to find proper-sized and irregularly-shapedtrackable features, the use of which can outperformprocedures using normal square features. The notionof confidence associated with each feature is introducedas the feature propagates. The use of confidence results inrobust tracking even when occlusion is present. Based onthe translational displacement of each feature, the affinemotion of the object can be accurately estimated. Thisapproach has been tested on a wide variety of videosequences and produces good tracking results.