Low-altitude road following using strap-down cameras on miniature air vehicles

Abstract

This paper addresses the problem of autonomously maneuvering a miniature air vehicle (MAV) to follow a road using computer vision as the primary guidance sensor. We focus on low-altitude flight with the objective of maximizing the pixel density of the road in the image. The airframe is assumed to be a bank-to-turn fixed-wing MAV with a downward-looking strap-down camera. The road is identified in the image using HSV classification, flood-fill operations, and connected-component analysis. The main contribution of the paper is the derivation of explicit roll-angle and altitude-above-ground-level (AGL) constraints that guarantee that the road will remain in the footprint of the camera, assuming a flat earth. The effectiveness of our approach is demonstrated through high fidelity simulation and through flight results.