The camera runs deep neural networks, after they have been trained on large sample data sets and uploaded to the Firefly. Running trained models directly on the camera should be able to produce on-the-spot image analysis for a variety of use cases, biometric recognition or quality inspections. By using deep neural networks rather than a rules-based object recognition system, the Firefly should produce more accurate results without having to align with explicit rules.
"The inspiration for Firefly is to subjectively analyze visual information," Mike Fussell, Flir's product marketing manager, said in a statement. "For example, it can inspect a manufactured part and identify defects that no one has ever seen before or even anticipated seeing. The result will be automation of visual tasks that previously could only be handled by humans."
During the early development phase, Flir used Intel's Movidius Neural Compute Stick (NCS) and Neural Compute SDK to prototype the Firefly. Intel launched the NCS, an AI accelerator on a $79 USB stick, last year.
Intel has been expanding its AI product line with different kinds of chips and products like the NCS under the mantra that there's no "one size fits all" solution for AI. The NCS makes it easy for startups and smaller firms like Flir to experiment with AI and, as Flir has done, expand their business into new areas.
The production version of the Firefly uses the Movidius Myriad 2 Vision Processing Unit (VPU) for image signal processing and open platform inference. It also uses Sony Pregius sensors and is compliant with the generic programming interface GenICam.
The camera is only about an inch square, it's small enough for handheld and embedded applications. It's equipped with a USB port for host connectivity, as well as four bi-directional general-purpose input/output (GPIO) lines.
The Firefly could be useful in a range of applications. In robotics guidance, for instance, it could help consumer robots interact with objects and avoid obstacles in unfamiliar spaces. It could be used for quality inspection purposes, to automate processes like looking for mistakes in textile manufacturing. Farms could use the Firefly to help analyze crop conditions, or medical professionals could use it for imaging implementations like fist-pass biopsy screenings.