Home -> Projects ->Local Feature Framework Member Login  

Jump to: Releases


In usual feature tracking system, generating local features consists of extraction intersting points and tracking them in successive sequences. Admittedly, feautre extraction and tracking has a close relationship, which means how to combine two modules effectively plays a key role on its comformance.

In our new local feature framework, it contains three major modules such as feature detector, matcher, and localizer. The demo videos show the results of the system in terms of tracking performance as well as correspondence mathcing performance using different feature extractor, descriptor, and matcher.

The system contains three major components
1. Feature detector using Harris Corner detection for extracting interest points and SIFT for generating feature descriptor
2. Feature matcher using distance between orientation distribution descriptor
3. Feature localizer using Normalized Correlation

 Demo Videos

Correpondent feature pair Matching Demo
The yellow lines show the correspondent feuture pairs between previous frame(left) and current frame(right)
Download: featureMatching.avi [28M]
Feature Tracking Demo
A movie shows the feature tracking result.
Red dots and green dots in left image show the detected features in previous frame and current frame respectively
Yellow lines in right image provide the tracking orientation. The length of each line doesn't reflect the real distance of two matched feature pair. Instead, it is magnified for user to see the performance
Download: featureTracking.avi [23M]

 Releases [ Members Only ]