I’m building a Python pipeline that aligns two images of the same sky area, subtracts them, and looks for moving-object candidates. I’m especially interested in ways to reduce false positives and what to do about objects whose motion is hard to detect from position changes alone.
How can I make this asteroid-detection pipeline more reliable?
Asked by
K6.Purlow
Answers
Before tuning the rejection rules, I’d check why the campaign uses its evenly spaced detections criterion. It may be there to reject image artifacts or require a track that is consistent with motion, not simply to exclude faint objects. Compare your rejected candidates against the original images and the campaign’s criteria, then see which failure modes your pipeline actually recovers.
Reply to this answer
Two-image subtraction is a useful starting point, but alignment and subtraction can leave lots of residuals around stars, bad pixels, and imperfectly matched image quality. I’d test the pipeline on known moving objects and on images with no expected motion, and keep intermediate aligned and difference images for inspection. A candidate that looks convincing only in the difference image may still be a subtraction artifact.
Reply to this answer
For the motion question, position-only detection has a real limit: if an object’s apparent position barely changes over your observation interval, this method may not distinguish it from a stationary source. More epochs and a longer time baseline can reveal a track that two images miss. Brightness or shape changes might offer clues, but they are not a dependable substitute for measurable motion on their own.
2N__Tumble:
Would measuring the point-spread function across several epochs help as a secondary check, or is that too sensitive to changing image conditions?
Reply to this answer
I’d be careful about inferring an object’s physical motion from pixel displacement alone. First calibrate the image coordinates to sky coordinates, account for the time between exposures, and track the uncertainty from alignment and centroid measurements. A short arc can support an apparent angular rate, but it may not be enough to classify what the object is.
Reply to this answer
Add your answer
Answers are reviewed before publication.
K6.Purlow:
That makes sense. I’ll try to separate cases where the criterion rejects a real-looking faint source from cases where it is doing useful artifact filtering.