From Sketch to Sensor: Synthetic Data and Commodity Hardware for Low-SWaP UAS Detection.
Counter-UAS is a crowded field, and if your use case and budget can tolerate an expensive, exquisite, or bespoke system, or one that is permanently installed or mounted on a vehicle or trailer, there are many to choose from. Our interest is the low-SWaP (size, weight, and power) end of the counter-UAS space: something small and light that a person can easily carry and set up in minutes, or already has in their pocket. Over the past year we have been exploring how much passive EO detection, tracking, and classification of small UAS you can get out of USB cameras, single-board computers such as the Raspberry Pi, and ordinary mobile phones. We wanted to understand what they can do, whether that holds up under real-world conditions, where their practical limits are, and how they fit as a layer in larger systems without depending on one.
This talk will demonstrate two threads of that work, both in progress. The first is synthetic training data: you cannot collect real imagery of every airframe you may need to recognize, so we render it, starting with 3D models of the airframes. We built those 3D models from photographs, technical drawings and blueprints, hand sketches, and plain verbal descriptions. All of them produced usable training data, and we will show what each gave us and what it cost. The second is the sensors themselves: fixed nodes built from a Pi, an AI accelerator, and a camera, plus an Android app that turns an ordinary phone into an acoustic-cued camera for detection and tracking. We will cover what we learned building and range-testing them, what we changed, and what we are taking back to the range in October.
About the Presenter: Dr. Kara Nance
To be added.
About the Presenter: Dr. Brian Hay
To be added.
