We have all seen that TV episode, from that law enforcement series you watch, where the analyst is looking at grainy security camera footage and the hero says “Enhance!”.
Like magic, the analyst draws a box around the car, types in a few keystrokes, and instead of a pixelated blob, you see it … a white Ford Explorer. Mystery solved, and the hero will make the world a safer place.
But that is silly, right? You can’t create new information. The original image didn’t have enough pixels to capture the detail to solve the mystery. Sure, you could interpolate between the pixels, but that would just make a smoother fuzzy blob of an image.
What if though, you could equip the hero with a superpower, call it super-resolution, that could let her enhance distant reconnaissance images? At STR, we’ve created that superpower.
We created a tool that can process the most distant kind of security camera footage — pictures taken from space — and uses artificial intelligence to understand what the object in the image likely is and then uses its existing knowledge of the world to guess what it would look like in more detail. Watching it work almost feels like magic.
“ONE OF THE GREAT THINGS ABOUT WORKING AT STR IS…MAKING SURE THAT THESE SUPERPOWERS ARE DELIVERED TO THE HEROES THAT ARE TRYING TO MAKE THE WORLD A SAFER PLACE.”
In more technical terms, we created a pipeline for processing satellite data that combines multi-stage image tiling and upscaling, a clever object detection architecture, and label stitching. One interesting fact that emerges is that a good AI pipeline can recognize that fuzzy blob is likely a vehicle, even though it is hard for a human. Then, using super-resolution, an algorithm can patch in detail that makes it recognizable for an analyst also.
One of the great things about working at STR isn’t just tackling hard problems like building an artificial intelligence system that can “Enhance!” on demand, but making sure that these superpowers are delivered to the heroes that are trying to make the world a safer place.