A Comparative Study of Two Approaches for UAV Emergency Landing Site Surface Type Estimation

摘要

An automatic landing site selection algorithm generates potential landing sites for unmanned air vehicles (UAVs) with engine failures. One important step in the landing site selection algorithm is surface type estimation. In this paper, we focus on distinguishing the following three surface types: grass/soil, tree, and inland water. Two approaches are presented. One is a conventional approach that combines Gabor features and a nonlinear classifier known as Support Vector Machine (SVM). Another one is a deep learning-based approach called SegNet. Extensive simulations showed that although both approaches achieved high performance, the Gabor/SVM approach yielded slightly better robustness with respect to illumination changes.