TOYON RESEARCH CORPORATION — Department of Defense SBIR Phase I: N231-056

Description

The U.S. Navy is adopting video sensors that produce ever increasing numbers of pixels at faster frame rates. To leverage these pixels, Artificial Intelligence / Machine Learning Algorithms are being trained and deployed to the video processing systems that support the sensors. However, to achieve robust performance of these and future algorithms the Navy must intelligently store a subset of the video data for future algorithm training without overwhelming the recording capacity of the deployed system. In Phase I Toyon proposes developing Deep Reinforcement Learning (DRL) agents to ingest streamed data and perform intelligent video recording that consider the size, saliency and quality of the recorded data. These agents will serve as the feasibility demonstration for the planned Phase II extension of these capabilities to consider complex video recording schemes and input data sources.