Request: change Hugin CP recommendations

[Klaus Föhl]

Hello,

For stitching panoramic images with Hugin, recommendations for placing the Control Points include:

1) place Control Points for all overlapping image pairs

2) spread Control Points over the overlap image area

From practical experience, results do improve, hence I suggest to change the recommendations as follows:

1a) place Control Points only for image pairs that are blended

2a) place Control Points only in the vicinity of the seam line

2b) if you move the seam line, also adapt your set of Control Points

2c) avoid placing Control Points into image corners

This request is based on stitching panoramics using images from more than a dozen cameras by more than half a dozen manufacturers, the number of panoramics being well into the triple figures.

The root reason is that with reference to the mathematically sound and physically motivated Brown-Conrady lens model only one distortion parameter is being used in hugin and panotools. Short of changing program code to include further (radial and possibly also non-radial) Brown-Conrady distortion parameters, the above recommendations mitigate the visual impact.

One might consider to change Hugin's CPFind behaviour of selecting CP placement areas. CPFind and Fine-tune in itself are ok.

I know I need to be careful not to be mislead by parallax effects and chromatic aberration. Based on mostly landscape panoramics my findings stem from

a) systematic inspection of the vis.*.tif files using --visualize in the enblend settings

b) detailed studies in several cases, observing image content shifts of several pixels, double figures even in corners

c) compass angle errors at the percent level when comparing hugin panoramics with synthetic panoramics from the udeuschle.de tool

Best regards

Klaus

[Bruno Postle]

I spent a few hours over the weekend watching someone stitching panoramas with ptgui. Tom was very concerned with only having control points on seam lines. He had two techniques:

1. He has written a tool that discards control points between non-adjacent images. This just uses the image order to determine adjacency, but I can imagine a better tool that considers likely seam positions and uses this to construct a graph of adjacent images, that discards connections between image pairs that overlap but which won't be blended.

2. Ptgui has a nice feature where the seam lines are drawn in the preview window. With two monitors you can use this as a reference for deleting points in the control point editor. It would be much more useful to actually draw any seam line between each pair of images in the control point tab (I know that Hugin can never predict where enblend will place a seam, but I assume that verdandi seams are similar enough to the unoptimised enblend seams).

--
Bruno

--
Bruno