This is a robust and accurate circle fit. It works well even if data
points are observed only within a small arc. This circle fit was proposed by V. Pratt in article "Direct least-squares fitting of algebraic surfaces", Computer Graphics, Vol. 21, pages 145-152 (1987). It is more stable than the simple Circle Fit by Kasa (file #5557).
Thank you very much for this algorithm. I was looking for an algorithm to fit a small number of data points (around 50) to a small portion of a circle (less than 10 degrees of the outline) and found this one to work perfectly. I had previously been using the Bucher circle fit and found it to be nowhere near as accurate or stable (although it is faster).