As far as I can tell, the methods used today are identical in spirit to what was done in the days of photographic plates, except that digital cameras and computers make all the steps a lot easier and less expensive.
The first method is blinking. You take several exposures, pausing for some time after each one, and then play them all like a movie. All the stars stay still because the telescope is following their motion. But if you are lucky enough to have a sufficiently bright asteroid in the field of view, it will show up as a little moving dot.
A related method is to take just three exposures, using first a red filter, then green, then blue. Combine all three to make a color image. The stars will appear white (more or less) while anything moving will appear as a sequence of three dots: red, green, and blue.
Finally, you could just take an exposure long enough that the asteroid makes a nice streak.
The streak method usually happens by accident, while someone is taking a long exposure for other purposes. It's not a great way to find new asteroids because:
- An asteroid has to be pretty bright to leave a beautiful streak like the one I have simulated here.
- You can't tell which way it was moving. ( Although you can probably make a high-probability guess. )
I don't want to do any of these things.
What I want to do is use image processing and machine vision techniques to allow us to find streaks that are really, really faint. So faint that they are right down in the noise.
This will allow us to do two cool things:
- Find rocks that are much fainter than the other methods can.
- Find them several times faster than either of the multi-exposure methods.
The concept is -- compute power is getting exponentially cheaper. Since I left grad school, compute power has gotten cheaper by a factor of 10,000. Telescopes have gotten more expensive. Moreover, compute power is continuing to get cheaper by a factor of two every couple years, while telescopes are continuing to get more expensive. If we can use compute power to make telescopes more effective at finding asteroids -- that would be a Big Deal.
But to do that, we will need to go to a cold and lonely place. The dark between the stars.
If you need an algorithm, here's one to consider: first, black out all of the point-source clutter (stars and unresolved galaxies). Then convolve the image with a 2-d matched filter for a short, fat streak along the x-axis. You'll probably want to do this convolution in the spatial-frequency domain, by multiplying point-wise the 2-d FFT of the image, and of the target object, then inverse-transforming. If the output of that filter is "significantly" greater in some spots than others, see whether the enhancement is point-like, or indicates a longer streak. Refine the detection with a set of narrower filters tilted over a small range of angles. (You pre-compute the whole set of transformed filters, so each filtering operation is mostly the inverse transform of the product.) Now that you've tested for streaks along the x-axis, repeat at non-zero angles until exhausted.
ReplyDeleteYou know, I think that's an awful lot like what I want to do.
ReplyDeleteIncluding the star-deletion.
I've already got the star-deletion part working, but I haven't gotten round to writing about it yet.