Skip to content

Scaling Moving Object Detection

Jump to bottom
Jeremy Kubica edited this page Nov 1, 2022 · 5 revisions

Scaling Moving Object Detection

Overview: The LINCC-Frameworks team is helping enable asteroid discovery and tracking on the LSST data by scaling moving object detection algorithms. The team's initial focus is the kbmod algorithm, which uses a shift-and-stack approach to find objects whose brightness might be below the detection threshold.

Code Repository: https://github.com/dirac-institute/kbmod

Issue Tracker: https://github.com/dirac-institute/kbmod/issues

Software Documentation: To be added.

Team Members: Dino Bektešević, Pedro Bernardinelli, Andy Connolly, Mario Juric, Bryce Kalmbach, Jeremy Kubica, Hayden Smotherman, Max West, Peter Whidden (in collaboration with broader kbmod team and community).

Internal Team Wiki: To be added.

Current / Recent Efforts:

The team's current efforts are primarily focused on enabling a v1.0 release of the kbmod software. This includes a range of code and project health improvements, addition of comprehensive testing, several new features, and the acceleration of the software. Specific efforts include:

  • Code productionization, testing, and health - Improve the overall code health and productionization of the kbmod software.

  • Algorithm acceleration - Improve the speed of the algorithm.

  • Scale the number of images - A current bottleneck for the kbmod algorithm is the number of images it can fit on GPU during a search. We (partially) removed this bottleneck by encoding images before sending them to the GPU.

  • Cross-CCD tracking (new feature) - Allow the discover of images whose trajectories span multiple CCDs.

  • Spatially and temporally varying PSFs (new feature) - Support the loading and usage of different PSFs for different times and CCDs.

References:

Clone this wiki locally