High Performace Template Tracking using Fixed Models

In this work we propose the graphics processing unit computational power to overcome a 2D tracking problem in a PF framework for more than real time performance (30 fps). We present a complete GPU rotation-translation-scale visual tracking solution that uses a previously acquired template for the evaluation stage of a PF. The key point is that the GPU offers hardware acceleration for interpolation and texture sampling that can be exploited by the application. It delivers much higher processing rates than interactive video playback enabling more robustness for the tracking problem. This GPU solution would be very promising when using future optimization strategies, such as adaptive appearance models for visual tracking.

Experimental results tested on an Nvidia QuadroFX5600 GPU exhibit impressive frame rates up to 220 fps for 640x480 video resolutions, 1024 particles, while successfully tracking a 64x64 patch. For the same PF and target configurations we obtain approximately 350 fps in 320x240 video resolutions.

Category:

Science & Technology

Tags:

• GPU
• real
• time
• tracking
• particle
• filter
• GPGPU

[multitouchy]

very nice!