Phase image-guided adaptive rotation-invariant feature point detector

Abstract

ABSTRACT Vision-based autonomous obstacle avoidance for unmanned air vehicle (UAV) is a vital research field to ensure the safety of UAV flight in the national airspace. In this paper, a proposed detector is introduced to extract the feature (corner) points of static obstacles from the surrounding cluttered environment for small quadrotor UAVs flying at low altitudes. The detected feature points are the bottleneck for most autonomous obstacle avoidance to facilitate tracking and identification of risk processes during UAV's take-off phase. The proposed feature point detector is based on the estimated phase image of the first image derivatives and the homogeneity test examination to give the proposed detector adaptive behavior. The performance of the proposed detector is compared with various feature point detectors in terms of several quantitative assessment indices and environmental conditions. From the experimental results, the proposed detector achieves promising results for significantly extracting the surrounding feature points.