[time-nuts] finding time astronomically.
albertson.chris at gmail.com
Mon Jan 23 19:21:00 EST 2012
On Mon, Jan 23, 2012 at 3:51 PM, Mike S <mikes at flatsurface.com> wrote:
> On 1/23/2012 3:02 PM, Jim Lux wrote:
>> How well could you do with something like the camera in the iPhone4
>> facing up. The front camera is VGA resolution
> A lower bound can be estimated.
> A cell phone (iPhone 4 rear camera) camera sensor has a resolution of what?
> ~2600 pixels wide with a 45 degree field of view - that's ~ 60 arc seconds
> per pixel, which is about 4 seconds of time. The Dawes limit is about 1
> second (17 arc-seconds) for a perfect .25" lens. Obviously worse with a VGA
> resolution camera.
The goal is not to create an image. A blur is actually better and
I've read of people intentionally using de-focus. What you do in
compute a best fit of the system point spead function (PSF). Or with
many blobs in the field you do a convolution of the image with the
The end product is not an image but a table of X,Y coordinates of each
detected star. You don't need to detect every star. Then you
search a star catalog and find thebest fit transformation matric that
takes you from X,Y to the catalog. The matric is your real
Typically you should expect about 1/10 of a pixel resolution at the
end. And then you take hundreds of images every night and average
them and you continue maybe for years.
If you were designing a camera for this purpose you make it so that a
typical star would cover maybe five pixels across so that the 5 by 5
pixel subimage would look like a Gaussian function. The centroid of
the function is your X,Y for the star. So you see that even with 5
pixel blurs you can likely find X,Y to much better than one pixel
width. This helps with noise too, noise would be a poor fit to a
2D Gaussian function. (and also there would be no catalog star for a
Redondo Beach, California
More information about the time-nuts