Attention-Based Vision for Autonomous Vehicles

The prime objective of this Phase II SBIR effort is to design and develop a novel computational visual attention framework for automated UGV systems. To achieve the program objective, Novateur Research Solutions and UCSD are developing biologically inspired computational visual attention framework that not only explains biological visual attention, but enables integrated bottom-up (unsupervised or data-driven) and top-down (task-driven or supervised) visual scene understanding. The framework will enable the onboard UGV perception systems to quickly focus on the task-relevant regions of the scene based on spatio-temporal context and the learned models of targets of interest. The Phase I effort demonstrated the feasibility and effectiveness of the visual attention framework to solve visual understanding problems on a variety of real-world data sets and laid a strong foundation for the subsequent work in the form of problem identification, solution development, performance evaluation, and detailed analysis. The Phase II effort will build upon the success of Phase I to mature and transition the technologies.