On Tue, Apr 12, 2011 at 7:28 AM, Florent Thiery <[email protected]>wrote:
> Hello, > > First, please let me introduce our use case; we are trying to use Elphel > cameras (353 + Computar 4-8mm 1/2") to get maximum resolution (> FullHD), 25 > fps video. > > Our current problems are about image quality when zooming on details > (blurred images); in other terms, we are trying to improve the rendering > quality as much as possible to have cleaner images. In this context, last > year we implemented http://code.google.com/p/gst-plugins-elphel/ but only > changing the debayering algorithm did not improve the quality enough for our > application (at least not when compared to the processing overhead). > > I was wondering about the method described in the awesome article "Zoom in > ... now enhance": > > - are there any specifics about the method being for Eyesis only (my > guess is it's not) ? > > Florent, sure it can be used with as single camera too. But it may be too slow for processing videos - it now takes 2-3 minutes/frame on an i7 with >8 GB of RAM (in multi-threaded mode) > > - regarding the calibration, what are the invariable factors ? Is the > calibration required for: > - every camera model/generation (depending on camera/sensor > manufacturing design/process variations) ? > - every lens model (depending on lens model) ? > - every lens tuning (zoom level / focus / iris ...) ? > - climatic condition changes (temperature, ...) ? > > This was designed for "fixed everything" (though we did not notice degradation with changing temperatures). But there is a significant difference between lenses even the same model. The lenses we used do not have zoom, and iris does not make much sens for such lenses - with 2.2 um pixels you can not close lens more than ~4-5.6 because of the diffraction, small sensors do not provide much control over DoF and using iris to limit light - not really needed with ERS sensors - they can handle very short exposures perfectly. And of course, focus setting would influence results too, as well as the lens orientation. > - the hidden question behind this is: how can this technique be used in > production ? > > Working with wide angle lenses ( 45x60 degrees FOV) we do not have enough room to capture the test pattern in a single shot, so software is able to combine multiple shots where the test pattern covers just a fraction of frame. We did not work on optimizing computational time of the calibration, so it takes several hours to process data ftom 8 camera modules. Current calibration is only designed for the aberrations correction, but I'm now working on the precise distortions calibration (with some 1/10 pixel precision), we plan to use it for panorama stitching, it can also be used for making measurements with the camera. This calibration will use the same pattern we use fro aberration correction, just at closer range, so the pattern will cover the whole FOV (minor out of focus is not a problem here). > > - For a given camera/lens combination, could a public database of > tuning data reduce the calibration requirement (in a similar fashion to > A-GPSes which download correction data from the network to increase > performance on low-quality reception and/or chips > http://en.wikipedia.org/wiki/Assisted_GPS) ? > > Our goal was to have precise individual lens correction, we did not experiment with correction of the lens model - it probably is possible, but with less correction, of course. Software has multiple tuning parameters, it should be possible to do that. > > - is there a hope of having such a feature (in the long term) > integrated in the camera itself (i.e. grabbing an mjpeg stream who had > the > corrections made right before the encoding) ? > > Not in the near future, at least. We now heavily rely on post-processing, camera role is just to capture all what sensor can provide, in as much raw form, as possible. Andrey > > Thanks > > Florent > > _______________________________________________ > Support-list mailing list > [email protected] > http://support.elphel.com/mailman/listinfo/support-list_support.elphel.com > >
_______________________________________________ Support-list mailing list [email protected] http://support.elphel.com/mailman/listinfo/support-list_support.elphel.com
