Dear List
Recently I have been reviewing various camera calibration methods with a
mind to creating a simple program that performs various functions specific
to different sport settings. While a number of very good commercial systems
exist, writing your own program allows the functionality of producing a
report that can be given directly to a coach. For example, a golf swing
input based on a template of markers and calculations could quickly return
information about pelvic and trunk rotations, club head speed and head
movement.
With that in mind I delved into a number of seminal works (e.g. Allard et.
al. (1997), Hinrichs et. al. (1995)) and reviewed several websites (Kwon3d,
Biomech-L archives) and to my horror discovered a whole world of camera
calibration techniques beyond the basic DLT I learnt as a student.
>From the Biomech-L archives I found the following suggested 'Golden
Algorithm' from Tomislav Pribanic...
> 1. Carry out DLT calibration
> 2. extract camera parameters with so-called qr decomposition of camera
> matrix (there is a paper explaining qr decomposition for the purpose of
> camera calibration, but I cannot think of it now; if you fail finding it
> also I'll try look for it)
> 3. Include non-linear distortion parameters and along with camera
parameters
I am the first to admit that my maths/programming is not strong enough to
quickly and reliably write anything more complicated than the basic 11
parameter DLT.
At what point does the extra complexity of calibration techniques stop
adding to 'useful' accuracy of 3D reconstruction? By the term 'useful' I
mean getting reconstruction accuracy to better than a few mm is perhaps less
of a concern than accurate digitising of body landmarks/joints or possibly
confounded by the accuracy of the control point survey.
With that in mind what would be the 'Golden Algorithm' for a sports
biomechanist who was happy with reconstruction error of around