Improved Scene Understanding Through Semantic Reasoning and Online Learning - Phase II

This SBIR Phase II project proposes the development of a robust and adaptable framework for scene understanding and false alarm mitigation that uses semantic reasoning to incorporate multiple sensors, external data sources, domain knowledge, and scene context. In the Phase II, Novateur will extend the Phase-I Markov Logic Network (MLN) based framework that exploits semantic reasoning and relational learning with deep convolutional neural network to remove high-confidence false alarms. The enhanced framework will include new MLNs with the ability to incorporate additional data types and novel contextual information. The proposed effort will build upon the vast portfolio of existing work by Novateur’s research team in the areas of target detection and classification, multi-sensor data and decision fusion, UGV perception systems, statistical relational learning, deep learning, and adaptive context-aware systems for scene understanding and inference.