Low-Shot Object Tracking and Targeting
Aided Target Recognition (ATR) capability for detecting and identifying targets in multi-channel EO and IR imagery.
LOTT is an AI/ML computer vision analysis framework optimized for the DoD mission space. LOTT includes multiple independent algorithm modules for target detection, identification, and tracking, as well as 3D scene reconstruction for hazard and obstacle avoidance. The target detection and identification algorithms leverage state-of-the-art AI/ML techniques, but adapt the underlying architecture to address challenges including small, obscured targets and to enable the use of multiple EO/IR input channels. Furthermore, we use self- and semi-supervised training techniques to enable the introduction of new sensors and target classes with minimal labeled examples. All LOTT capabilities have been demonstrated to execute in real-time on embedded compute platforms integrated with mobile systems. The algorithms scale across multiple processors for a several tasks, including analyzing single long-range sensors at distances up to 5 km and providing a complete situational awareness display across a full suite of cameras for 360 degree coverage around relevant vehicles.
- Fusion of multiple EO/IR imagery to maximize performance in all lighting and environmental conditions
- Uses self- and semi-supervised learning techniques to fully initialize deep neural networks with minimal labeled data
- AI/ML architecture improvements to enhance detection of distant targets and provide robust identification
- Executes in real-time on embedded hardware on mobile platforms
Benefit to Warfighter
ATR algorithms reduce the warfighters’ cognitive burden by continuously monitoring high-clutter scenes for new threats and rapidly directing soldier attention to potential threats when present.