Discipline: Computer Sciences and Information Management
Subcategory: Computer Science & Information Systems
Andrew Morgan - Youngstown State University
Co-Author(s): Zachary Jones, Marshall University, Huntington, WV
In order to maintain safe flying environments and avoid disastrous mid-air collisions, the U.S. Federal Aviation Administration mandates that pilots must ‘see and avoid’ other aircraft. The purpose of this research is to develop a See-And-Avoid (SAA) system for autonomous Unmanned Aerial Vehicles (UAVs). In this paper, we present a novel approach to the development of SAA capabilities using simulated cockpit video and computer vision algorithms. The cockpit video is generated using the computer graphics library OpenGL. The computer vision library OpenCV is used to construct algorithms for background filtering, obstacle detection, and collision avoidance. We hypothesized that through the utilization of computer vision, we are able provide robust algorithms for SAA situations of head-on, mid-air collisions. This research is pivotal in unmanned aircraft research; no other author has utilized a custom flight simulator to provide applicable data to mid-flight collision avoidance. Through this work, we are able to provide an approach necessary for continuation of research in unmanned, autonomous flight. After construction in the OpenGL graphics library, a flight simulator was developed to exemplify real-life flight situations. Through this simulator, we had the opportunity to add and subtract 3D components, providing an influential test bed for collision avoidance algorithms. All data collection was evaluated through algorithms built into the OpenCV computer vision library, which controlled object detection and object placement. This library presented predefined functions essential for proper object detection, including edge detection, blob detection, and transformations. It was furthermore necessary to track potential threats over several camera frames. Through design of a unique Object ID tracker, it was possible to further examine the history of threats over time. In evaluating threat levels of objects, two unique algorithms with associated avoidance maneuvers were constructed and tested. As control components, situations of collision were processed to compare avoidance algorithms with no avoidance at all. The SAA algorithms provided a great deal of security and protection to the unmanned aircraft. In avoiding all situations of direct collision visible to the camera, each algorithm possessed unique benefits to the avoidance maneuver. Results from the tests are furthermore discussed in the scope of this paper. Future work will focus on improving the ability of the SAA system to identify obstacle aircraft type, direction, and speed. Using this information, sophisticated avoidance algorithms could be used to minimize collision rates and maximize efficiency.
Funder Acknowledgement(s): National Science Foundation Research Experiences for Undergraduates at Auburn University 2016
Faculty Advisor: Richard Chapman, chapmro@auburn.edu
Role: My work with this research specifically pertained to the computer vision associated with collision avoidance. Working alongside my research partner Zachary Jones, we were able to work cohesively in the development of computer vision components. Afer we developed the OpenGL flight simulator, we began constructing algorithms in OpenCV to detect and analyze objects in the aircraft's field of view. From detection of potential threats, avoidance maneuvers were developed according to threat levels determined by our computer vision algorithm. By tracking threats over a given number of frames (30 frames), we were able to more precisely develop an algorithm for enhancing collision detection procedures. Avoidance maneuvers were generated according to the history of the threat.