Object Counting using Unified LAtent Representation (OCULAR)

Vadum will develop and implement a novel, unsupervised deep-learning approach to estimate the number of objects in a SAR image automatically, accurately and with low-latency. The approach learns a unique representation of a SAR image that is resilient to a wide range of SAR artifacts, such as geometric and temporal image misalignments, resolutions, noise and collection geometries. The technique architecture is extensible to fusing images collected from other radar imagery sensors, as well Infra-red (IR) and Hyper-Spectral Imaging (HSI) sensors enabling high quality, all-weather data collection capabilities. It can be deployed in a number of scenarios: An analyst can depend on automatic, accurate counting estimates and focus on complex operations such as scene situational understanding; At scale deployment in distributed environments simultaneously covering large swaths of geographical areas; Standalone deployment in edge environments such as on-demand scene understanding for remote piloted aircraft.