Finding moving objects below the single-image detection limit
Abstract
Title: Finding moving objects below the single-image detection limit We present new results from the Kernel Based Moving Object Detection (KBMOD) pipeline, which detects moving objects that would be too dim to be detected in a single image. This pipeline takes a series of images of a given field, creates matricies of the likelihood of a source detection in a given pixel, and uses GPU-accelerated computing to "shift-and-stack" these likelihood matricies along a grid of position and velocity vectors, returning stacked maximum likelihood results over a given threshold. We then use a lightcurve filter followed by a image central moment filter to help eliminate false positives. Unlike traditional linking algorithms which require source detections in a given image before linking can occur, KBMOD does not require a source detection in any individual image in a given field. Furthermore, KBMOD does not need any fixed image cadence in order to detect moving objects. The result is that we are able to use consumer-grade hardware to quickly search a stack of image and detect moving objects that may be too faint to be even detected as sources in a single image. Here, we follow up the work of Whidden et al. 2018 by applying KBMOD to a new 6.5 TB data set covering over 2000 square degrees of the sky. We now use difference imaging software from the LSST Software Stack on the data set prior to running KBMOD. This not only gives us a new data set in which to search for moving objects, but also allows us to avoid simply masking sources out of images, thereby increasing our effective searchable area for a given image.
- Publication:
-
American Astronomical Society Meeting Abstracts #233
- Pub Date:
- January 2019
- Bibcode:
- 2019AAS...23325506S