Tracking in realtime is a challenging task. With the interactive TemplateTracker we offer a small and efficient C++ library which leverages the power of Nvidias Compute Unified Device Architecture. Basic element of the library is the template matcher component, which is the processing backend running on the GPU.

To use it, we have made several frontends:

• A Qt frontend for Nokias framework library and its QImage format.
• A frontend for OpenGL based image buffer formats.
• Additionally, for Intel IPL/IPP users, an IplImage based frontend is planned.

Potential application areas:

• Track an object to get the trajectory of it.
• Tracking of multiple points of an object to figure out camera zooms, pans etc.
• Detection of known, fixed size objects, e.g. a logo.

Technical details
The number of tracking points is unlimited. Blocksize and tracking region of interest (ROI around the current block) depends on the compute capability of the graphics card. Blocks and ROI dimensions can be adjusted in case you need rectangular blocks rather than square ones.

The TemplateTracker library is available on Windows and Linux platforms.

Demonstration Application
A small demo application can be downloaded here.
It is tested with imagesizes up to 2K (2048x1535). To run it, NVidias CUDA SDK 2.1. or 2.2. and an appropriate graphicsdriver must be installed. You must provide your own image sequence as input.
For further information on the operation of the demo apllication, please refer to the readme.txt inside the demo package.