Often I get the urge to dive into a dataset and just start having fun. That was definately the case with the Sloan Digital Sky Survey (http://www.sdss.org).
I wanted to try out some of the newer features of opencv (http://opencv.org) combined with modest genetic algorithm approaches in the detection of moving objects wihin SDSS images.
SDSS uses ccd (charge coupled device) camaras for imaging the night sky. The red, green, blue channels are scanned slightly offset in time, which means a moving object can easily be observed.
A few stolen snatches of time and I have come up with a first naive attempt:
Which will mark out a potential asteroid with a circle. For now I am dumping the candidates to an Amazon S3 bucket http://sdss.asteroid.detect.s3-website-us-east-1.amazonaws.com/
I am now working on a more sophisticated approach … which will take care of those pesky false positives and attempt to rank an image quality first before I run detection algorithm. My initial investigations were to familiarise myself with the dataset. I like working with significant subset of a dataset before I start enshrining design decisions.
Will keep you posted …