Realtime image registration stacking for live viewing!

[NickK]

I'm on vacation so I've spent a day writing automatic alignment of images (that rotate, scale and move) so that images can be stacked in "real time".

This means you can use your camera like a video camera, each frame taken (say 5 second exposures) is aligned and stacked automatically in realtime.

The code isn't finished, but currently will automatically detect, align and stack 17MB 383L frames in 8 seconds with lots of diagnostic images being generated during that time too. I should be able to get this down to 1-2 seconds which, when coupled with a 5 second exposure will result in a near realtime "live view"

Below is a scaled down example rotation of the frame - the software detects the misalignment and corrects automatically .. in the next post is the correct image.

[NickK]

The attached image has been automatically corrected.. a little more work and it will sit within the Example Application - alongside the "static stacking" which doesn't perform the alignment.

[ForumAdmin]

cool,  I have enjoyed doing this kind of thing with webcams at star parties in the past.
Steve

[NickK]

An update - I've improved the accuracy with interpolation. The result is it's doing a very good job of alignment; actually it's getting accuracy to 3 decimal places to what pixinsight gives!

Still lots more testing and playing - it's very much development at the moment but the idea is it will be added to the example application

Nick

[NickK]

Native resolution alignment - images added together:

[NickK]

Here's an 800x600 image of the two aligned images, showing the overview images:

[ForumAdmin]

Would I be right to assume that the camera was not tracking?  The 383 is capable of drift scan imaging but we never implemented this.  Might be a project for a rainy day
Steve

[NickK]

Hi Steve,
The mount wasn't aligned at all - literally placed down pointing north but set to track at sidereal rate (no guiding). This meant that the few seconds for exposures didn't cause trails but the target drifted over time (in this case it's saturn at the bottom as an easy target to use for testing).

Drift tracking would be easy to implement. The main thing is getting the alignment speed correct - and using optic flow the mount tracking speed could work for off the sensor without needing PHD for example. This conversation sounds familiar but the shutter would need to be kept open during the continous reading

Nick

[NickK]

I have to add that deliberately didn't have the cooling running to get as much noise as possible as a test case.

[NickK]

Here's an additive stack of the two input images.

Here you can see what would happen without alignment and the images were stacked "statically" - if your mount is aligned perfectly you can do without alignment and just stack but most mounts can't get the accuracy.

[NickK]

Youtube video of the Example App aligning and stacking

http://youtu.be/yOoAJDAJ9xM