Turbulence not only destabilizes aircraft flight, but can also affect aerial photography systems used in surveillance, astronomy, etc. Long-distance photography is particularly challenging, as the airflow between the camera and the subject is disrupted, often distorting the photos and making them nearly useless.
Experts in machine learning and computational image processing at the University of Maryland have received a $360,000 grant from the Army Research Office (ARO) to tackle this challenge. Christopher Metzler, an assistant professor of computer science with an appointment at the University of Maryland Institute for Advanced Computing (UMIACS), received the funding through ARO's Early Career Program.
The grant will support a three-year project to develop a system that uses high-speed cameras and machine learning algorithms to instantly produce clearer images, even in areas with extreme turbulence.
“No matter how good your optics are or how much money you spend, you're still fundamentally limited by the atmosphere,” Metzler said.
Current approaches to long-range imaging focus on measuring the level of distortion and then gradually adjusting the optical system to correct for atmospheric interference, he said.
But these types of systems are hard to use for capturing fast-moving objects. To overcome this, Metzler and his graduate students are using neuromorphic cameras (also called event cameras) that are triggered by movement or changes in overall intensity to capture data under a variety of conditions.
Conventional cameras used for long-range imaging can capture images at a rate of 100 frames per second. Neuromorphic cameras can capture 10,000 frames per second with relatively little power.
Metzler plans to design new machine learning algorithms to process this large amount of non-traditional image data, as well as proprietary optical hardware that will enable it to obtain more diverse measurements that are packed with more information.
One of the team members, doctoral student Sachin Shah, is lead author of a paper published recently at the IEEE/CVF Computer Vision and Pattern Recognition conference in Seattle that describes some of the strategies the researchers plan to employ.
