Visualizing Behaviors when Using Real vs Synthetic Imagery for Computer Vision

摘要

Unmanned aerial vehicles have become a ubiquitous tool for humanitarian and intelligence operations world-wide. However, in-flight refueling cannot be safely conducted because of high latency inherent to the command & control signals. One approach to over come this is to equip the refueling tanker with a stereo vision system that observes and computes the pose of an approaching receiver. To test this hypothesis, we virtual environment that produces synthetic imagery of an approaching receiver. Similarly, we also created a real-world physical replica of this approach. This enables us to directly compare virtual imagery against actual imagery to test the fidelity of synethic stereo imagery for relative formation flight. This work employs a virtual environment to generate synthetic imagery of an aerial refueling scenario involving small a quad copter. This research visually compares the performance of stereo vision algorithms on synthetic and real world imagery.