Our video registration dataset is available for researchers to use for the comparison of registration techniques. Any interested researcher should contact us for more information on how to obtain the dataset.
The dataset contains video clips from two football games. Each video clip has a corresponding file containing hand-labeled ground-truth mappings from landmark points in selected frames of the video to their corresponding points on the NCAA football field model depicted below and included with the dataset. Also included is the set of reference images we use in the experiments described in our paper. This set is also depicted below, hand-registered to our field model. In addition, we include C source code for a function to import data from a ground-truth registration data file and a sample program demonstrating its use.
Our 2D model image of an NCAA football field
Our set of reference images hand-registered to our field model
File Format
For more information on the registration ground-truth data file format, see this page.