Collaborative Research in Multi-Target Sense and Avoid for Small, Lightweight Unmanned Aerial Vehicles (UAVs)
Abstract
Over the past year, researchers at the Naval Postgraduate School (NPS) and Purdue University have worked collaboratively to address fundamental research problems in the automatic detection and tracking of unmanned aerial vehicles (UAVs) from high definition video cameras on other UAVs. This technology, known as visual sense-and-avoid, is of crucial importance for coordination and situational awareness when many UAVs are present in the same airspace. Our research effort has resulted in the submission of a conference paper [1] which lays out a novel approach to the problem of real-time on-board sense-and-avoid technology for small UAVs. While the detection of moving objects has been an important problem in computer vision for many years, the problem of visual sense-and-avoid for UAVs presents the following important new technical challenges. • Small targets: The target aircraft being detected are often small and distant meaning that they occupy a very small portion of the field of view; • Background clutter: The targets are typically moving in front of a complex clutter background that can obscure targets; • Camera motion: The background image has a great deal of apparent motion due to the rapid and unpredictable motion of the UAV camera platform; • Robust detection: Detection tracking must be robust to a wide range of variable lighting and camera distortion conditions; • Low latency: Low latency detection and tracking is required for closed loop control and decision making; • Light-weight on-board computation: All computation must be done onboard the UAV with low weight and power systems. The technical approach proposed involves advanced image processing and computer vision techniques for distant aircraft detection. Solving such a difficult problem requires integrating spatial and temporal cues using models of uncertainty in order to make robust decisions. Our proposed solution discriminates between target aircraft and false alarms by modeling the background motion together with the target aircraft’s morphology and kinematics. The video stream is processed in a pipeline architecture that integrates information at each stage using a model of uncertainty for background motion, target motion, and target features.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Sep 20, 2016
- Source ID
- N002441620003
Entities
People
- Juan Pablo Wachs
Organizations
- United States Navy
- University of Virginia