NavList:
A Community Devoted to the Preservation and Practice of Celestial Navigation and Other Methods of Traditional Wayfinding
Re: Star CN with DSLR Camera
From: Peter Monta
Date: 2016 Jan 27, 20:40 -0800
- The overall idea, of course, is to estimate the Moon's position among the stars as a function of the local clock, then solve for the time error. It's difficult to do this with a single photo because of dynamic range issues---the Moon is much brighter than the reference stars, so the detector saturates, obscuring the exact location of the lunar limb; also atmospheric scattering is strong at these glancing angles, injecting noise into the star background out to at least several degrees from the Moon's center.
- On land, stable platforms are available. Even the humble surveyor's tripod, or even humbler photo tripod, is surprisingly good. A couple of months ago I did a run of exposures near the zenith on a photo tripod under good conditions (no wind) for another project. After pushing the images through the astrometry pipeline, then plotting the declination of the center point of each image, the scatter was 0.4 arcsec RMS, and drift over the ten minutes even less. (I can make these images available if there's any interest.) On a surveyor's tripod or a concrete pillar it might be even better. Use an intervalometer box so you're not touching the camera; also use the special modes on the Canon cameras that keep the mirror up and use electronic first shutter so there's no vibration from the shutter either.
- So given that the tripod can accurately maintain the camera's direction, one can make two exposures, one long to capture reference stars and one short to capture the Moon's limb. Timestamps for each exposure are required. In practice one would do a long chain of exposures, alternating between short and long.
- Lunar limb or lunar features? I go back and forth on this. (Need to do more and think less.) On the one hand, the limb is a very distinct, high-SNR feature. On the other hand, the limb has topography (mountains) which might be a hassle to ray-trace out (although I'm sure good digital elevation models are available); also the entire limb is not present, resulting in bias worries. The other tack is to use a collection of craters, or, for more statistical power, the entire lit surface suitably warped. The libration models are so good that all of that can be predicted to far better accuracy than needed.
- Field of view and plate scale. I'm not sure what the optimal field of view might be. Larger fields capture more reference stars for astrometry, but each pixel is larger on the sky, hurting positional accuracy. That relationship is not direct, though. The papers say that typical uncertainties, even with undersampling, are on the order of 0.005 pixel using stars with high SNR and for which the system PSF is well characterized. There's also the issue of the pixel response function with a subsampled detector like a consumer DSLR. With the green pixels in a quincunx pattern the spacing goes up by sqrt(2), but also, even worse, the pixel response function is now a strange checkerboard thing, very unlike the usual uniform, square astronomical CCD or other array detector. I'm kind of waiting for the astrometry centroid models to get smarter about this. Existing estimators are unbiased, maybe, but noisier than they should be.
- Camera lens, distortion, calibration. All other things being equal I think I agree with Greg Rudzinski that zooms are a little riskier. They're more mechanically complex and might flop around more as the gravity vector changes. For either zoom or prime, though, you're probably going to have to estimate the distortion for each session. Best to point the camera at an empty stretch of sky at the same topocentric altitude as the Moon, then take a bunch of images and estimate the distortion field from those. The refraction component of the distortion will then be almost the same as the images with the Moon. Given that the Moon exposures will be short, with few reference stars (maybe 20 or 30), it's best to fix the distortion estimate prior to the Moon images. That way you're estimating only position and rotation, with distortion already taken care of.
Hope this helps.
Cheers,
Peter
From: Peter Monta
Date: 2016 Jan 27, 20:40 -0800
Hi Bill,
It's a good project with plenty of interesting overlap with astrophotography and astrometry. I've taken a few passes at this, as have other NavListers, but have yet to really equal or exceed the Markowitz effort :-). (He used a large telescope and a fancy "dual-rate" moon camera back in the late 1950s.) Here are a few ideas, assuming all of this is happening on land.- Filters or masking. I've tried suspending an ND3 filter (1000x) a few meters in front of the camera, in line with the Moon. It works but it's a hassle---must repoint every minute or so. Also you'd need to rotate the filter by 180 degrees between exposures to debias it.
- Timing. I tend to take photos of my phone's screen running an NTP app (0.1 second resolution). These photos are then used to estimate the camera's clock bias, which is then added to the timestamp in the EXIF header. A little GPS gizmo with non-multiplexed LED display might be better.
- Image format. It's essential to use RAW rather than JPEG.
- Software. For coarse blind estimation, astrometry.net. For fine-tuning, the tools from Astromatic.
