Moving Object Detection Software

In our Aug 1997 survey, we collected about 2 gigabytes of raw data per night, covering about 4 degrees to limiting R magnitude of 22.9. The tried-and-true method of finding slow moving objects (such as KBOs) is to take three pictures of the same field of sky. These images are then displayed in sequence and moving objects are found "by eye". This method (called "blinking") is quite effective, but very time consuming (add anim). It might take an efficient person 15 to 30 minutes to blink one of the eight chips of our 8k CCD. That breaks down to about 90 hours of blinking per night's worth of data collection.

The problem is that a KBO must be re-observed ("follow-up") within a few nights of its initial discovery or there will not be enough information about its orbit to predict its future position in the sky. Ideally, one would like to find the objects in real-time. So that objects found one night can be followed-up the next.

We have created a computer program called Moving Object Detection Software (MODS) to do the most of the analysis for us. We started from scratch and made the simplest, fastest, most robust program possible. It processes the data by identifying potential KBOs and circling them in the data (this takes about six hours for a night's worth of data). The observer then displays the images in question, much like the traditional "by-eye" method, but this time only the circled areas are examined. Instead of about 90 hours of blinking time per night, the workload is reduced to about 3 hours. You can see why we call this type of analysis "speed-blinking". (add anim)

Click here for a full-color illustration of MODS in action.

How well does MODS work?

One might ask how well MODS works. This is actually three questions in one.

• How well does it find moving objects, or what is its efficiency? We have done extensive tests of the program's efficiency at finding different brightnesses of KBOs. MODS finds about 95% of bright KBOs, and towards the fainter end, efficiency decreases as faint objects become lost in the sky background. Since the program efficiency is a function of object brightness, if one characterizes program performance well enough, it is easy to deconvolve the detected KBO brightness distribution (which will always be deficient in bright objects) and produce the actual KBO brightness distribution.
• How many false detections are there? Any computer program will not analyse data perfectly, and it's much better to have your program over-sensitive (that is, it detects almost all real objects but produces some false detections) than under-sensitive (it misses detecting real objects but produces no false detections). The number of false detections is typically a few per CCD per field. This means about 1000 false detections per night in our Aug 1997 survey, but it is relatively easy for the speed-blinker to reject most of these. Typical sources of errors are cosmic ray hits and background noise bumps that mimic KBO movement and diffraction spikes from stars.
• How robust is it? By "robust" (in this case), I mean "insensitive to operating parameters" which is akin to "will it break". The answer is no. We have tested the performance of MODS over a wide range of operating parameter values, and we find very little efficiency loss.

For a more detailed (and quantitative) description of the program's performance, and its application to a recent survey, see our upcoming paper, Trujillo and Jewitt, accepted in the Astronomical Journal.

Click here for a reprint.

Chad Trujillo
email: trujillo at gemini dot edu