|
There are large amounts of image data lying around the place in
various formats and in various stages of processing, and yet every
time I want to test a parameter estimation algorithm I seem to need
to take more pictures. The kind of data we're talking about is images,
feature points and lines detected in the images and other information
that meay be usefull in, or created by the process of estimating
the relationships between the images. So Fundamental Matrices, Projection
Matrices, Trif-focal Tensors and the like, but also principal points,
focal lengths, reconstructed scene points and the like.
In order to alleviate this problem whereby data is constantly having
to be regenerated I'm in the process of defining an XML
data type definition that will allow existing data to be stored
in a consistent form so it can be re-used. A subsidiary goal is
to set up a standard set of test images (and point data) against
which estimation methods can be tested.
The advantage of XML is that it's not platform dependent, it's
extensible (and so relatively future proof), it's not all that hard
to process, and it can be displayed on the web easily (so we can
see what we've actually got).
The files
|