APPLYING THE MULTI-REFERENCE DETECTOR TO COMPUTER VISION PROBLEMS
Matthew Schubert, Research Scientist, NIA
April 1, 2015, 1:00 pm, NIA, Room 101
Abstract:
The Multi-Reference Detector uses feature-based matching to find matches from a query frame to multiple different reference images. It can then localize the references in the query frame and, if provided a probability model, display probabilities based on whether specific references were found in the frame. It can been applied to computer vision tasks including runway identification and astronaut helmet detection. Both applications will be discussed.
For runway identification, the Multi-Reference Detector was used to help create a probability model relating images from runway approaches with their respective runways. Then, when providing the images and the probability model, the detector could estimate what runway a video or camera was showing at any given frame based on detected images. This detection is both specific, with a low false positive rate, and repeatable across approaches to the same runway in similar weather and time of day conditions. It has also shown some repeatability between visible and short-wave IR sensors.
For astronaut helmet detection, the detector can be used to locate the helmet in a camera frame when given images of “feature-rich” patterns affixed to the helmet. Based on which patterns were found, the relative orientation of the helmet can be determined for a wide range of viewpoints.
Bio:
Matthew Schubert received his M.S. in Computer Science and his B.S. in Information Systems from Christopher Newport University in 2011 and 2014 respectively. He currently works in the Computational Vision Lab at NASA Langley as a Research Scientist for the National Institute of Aerospace.