This research is investigating the feasibility of using computer vision to provide robust sensing capabilities suitable for the purpose of UAV collision avoidance. Presented in this paper is a preliminary strategy for detecting collision-course aircraft from image sequences and a discussion on its performance in processing a real-life data set.
Initial trials were conducted on image streams featuring real collision-course aircraft against a variety of daytime backgrounds. A morphological filtering approach was implemented and used to extract target features from background clutter.
Detection performance in images with low signal to noise ratios was improved by averaging image features over multiple frames, using dynamic programming to account for target motion.
Preliminary analysis of the initial data set has yielded encouraging results, demonstrating the ability of the algorithm to detect targets even in situations where visibility to the human eye was poor.
Source: Queensland University of Technology
Author: Ryan J. Carnie | Rodney A. Walker | Peter I. Corke