PILOT AI LABS, INC. — Department of Defense SBIR Phase II: AF191-005

Description

Pilot AI will develop a computer vision algorithm that quickly detects vehicles within very large format images taken from scanned Optical Bar Camera (OBC) film. Automating this process through the use of a fast computer vision algorithm will allow the 548th to better prioritize images for analysis, and highlight areas of interest for analyst prioritization. The model will be trained to be tolerant of significant optical aberrations or distortion present in high-altitude imagery. Pilot AI will utilize our tools and software developed commercially for the adaptation of US Air Force (USAF) use cases. These tools include: -Pilot AI’s proprietary computer vision framework for training and running fast neural networks for object detection -Neural network model architectures designed to run on ISR imagery as well as fisheye (i.e., heavily distorted) images from security cameras Pilot AI’s objective is to develop a model whose accuracy and processing speed are both high enough to be useful to the 548th. Pilot AI will work with the 548th and AFRL to determine inference speed as well as accuracy thresholds; Pilot AI will work in concert with AFRL throughout Phase II to measure key metrics of model accuracy/performance including F1, precision, and recall.