Integrating Compressive Sensing & Machine Learning for Outer-Loop Target Tracking Control on an Autonomous Quadrotor Aircraft
Abstract
Recognizing, tracking, and intercepting moving targets is a central mission of robotic surveillance and guided munitions. Automated tracking of a dynamic target poses a set of unique challenges, including motion, time-varying geometry, partial occlusions, variable illumination, and potential handoffs between multiple mobile sensing platforms. Currently, vast amounts of sensor data are being collected for vision-based target tracking and control. Only a small portion of this data is of use at any given time; a majority of the information is discarded by signal compression and the remaining data is further reduced for feature-based recognition. In this grant, we are building on the theories of compressed sensing and machine learning to develop an integrated framework for classification using a few key measurements. Our goal is to advance these algorithms such that they are robust and efficient enough to be implemented for autonomous real-time target tracking on a quadrotor aircraft platform in a future flight demonstration effort. We propose to combine our recent work in sparse sensor placement for decision-making with automatic target recognition and dynamic target tracking. In addition, we will explore algorithmic extensions that leverage streaming feature extraction to increase efficiency in dynamic scenarios.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Mar 10, 2016
- Source ID
- FA86511610003
Entities
People
- Bingi W. Brunton
Organizations
- Air Force Research Laboratory
- United States Air Force
- University of Washington