CAI: Can you spot the falling debris ?

Back in 2003, during its launch, Columbia's side wings were hit by items falling off the main booster. At 500 km/hr, the items literally broke the fragile edge of the orbiter leading it, fifteen days later, to its demise. When the mishap commission looked into the issue of falling debris at launch, it found several incidents over the course of the whole space shuttle program starting at STS-1. One of the remedies for the remainder of the program was to have cameras installed on the boosters and other locations on the orbiter during launch. It also led to the orbiters having to perform the Rendez-Vous Pitch Maneuver (RPM) thereby allowing  astronauts on the International Space Station to photograph every parts of the leading edge and the belly of the orbiters to confirm that no debris had in fact breached the hull.

Today's Cable And Igor's adventures in the evaluation of Matrix Factorization for Images and Videos (CAI) is focused on evaluating the Robust PCA as implemented on GoDec in some of the videos taken by cameras installed on both rocket boosters of the space shuttle. The video was taken on Atlantis during STS-135 the final mission of the shuttle program. Here are some of the Robust PCA analysis of the launch from different angles and cameras.. Can you spot the falling debris ?

It looks ok but Cable and I could not figure out why we were missing the big dark debris at the very beginning. It turns out it was there but we could not see it. The sparse and noisy part are too dark because we used Matlab's imshow function (the debris was just above the background but not enough to be bright on our screens). To remedy this, we used Imagesc instead which will give some better contrast

Here is another excerpt from the other booster
 

Sure, we could perform all kinds of dedicated image processing to locate the debris image by image, but what is very natural with Robust PCA is that it doesn't require an image processing background and it has the ability to automatically decompose the interesting stuff (the sparse component in this case) from the generic image (the low rank part) and any of the atmospheric elements (the noisy part). We like the fact that the noisy component looks like the result of some computational fluid dynamics simulation. Another item of note is that specular reflection on the wings is represented in the sparse component.

We took excerpts from this video:

Credit video: NASA

Liked this entry ? subscribe to Nuit Blanche's feed, there's more where that came from. You can also subscribe to Nuit Blanche by Email, explore the Big Picture in Compressive Sensing or the Matrix Factorization Jungle and join the conversations on compressive sensing, advanced matrix factorization and calibration issues on Linkedin.