GPU accelerated object tracking. Achieved speed-up: 40x

- Tracking -

(Single and multiple object tracking in HD video streams with GPU acceleration)

The Explosion Research Institue has worked on using the GPU to accelerate the
tracking of one or multiple objects. In this project, video data showing living
nematodes (roundworms) had to be analyzed and evaluated.


Our tracking software allows to speed up the task of tracking objects in a video by at least one magniture in comparison to conventional tracking techniques. Using conventional methods, tracking is achieved in a two step approach. First, image data is obtained from blue-ray video as raw bmp bitmap files. Then, in a second step, the bitmap images are analyzed using Matlab. This process is  accelerated dramatically by using our proposed tracking system. An outline of the features is given below:

・ 40x speed-up over conventional tracking (TMPGEnc + MatLab)
・ Processing speed: >100 fps for HD movies (1920×1080) on a recent CPU & GPU.
・ Direct GPU accelerated video analysis without  storing intermediate image files
・ Support for most video formats, including HD video formats (.m2ts , .vob)
・ GPU accelerated detection of objects in the video (using CUDA)
・ Fast multi-object tracking
・ Multi-core 64 bit  video decoding
・ Windows 32/64 bit support
・ MS Excel  and Open Office Calc compatible tracking data output (.csv)

We have implemented the tracking software using MS Visual C++ 2008. It is optimized for 64 bit and uses the Direct Show API for video decoding. The visualization is optional and implemented using the OpenGL API.

Development MS Visual  C++ 2008
Video API MS Direct Show
Graphics API OpenGL

NVIDIA CUDA is an API that allows to execute conventional C++ code on the GPU


Download movie of the entire simulation, WM9 compressed:
High Quality (26MB) Medium Quality(9MB) Low Quality(7MB)

  • Please contact us with any questions or comments that you may have.