An effective approach to unmanned aerial vehicle navigation using visual topological map in outdoor and indoor environments

Abstract

Unmanned Aerial Vehicles are constantly being using in professional activities that require higher precision in navigating and positioning the aircraft during operation. Advanced location technologies such as Global Navigation Satellite System and Real-Time Kinematic are widely used, however, they depend on an area with transmission coverage. In this approach, this article presents a visual navigation methodology based on topological maps. We compared the performance of consolidated classifiers such as Bayesian classifier, k-nearest neighbor, Multilayer Perceptron, Optimal Path Forest and Support Vector Machines (SVM). They are evaluated with attributes returned by last generation resource extractors such as Fourier, Gray Level Co-Occurrence and Local Binary Patterns (LBP). After analyzing the results we found that the combination of LBP and SVM obtained the best values in the evaluation metrics considered, among them, 99.99% Specificity and 99.98% Precision in the navigation process. SVM reached 5.49787 s in combination with LBP completes the training in 5.49787 s. Concerning the testing time, SVM achieving 80.91 ms in association with LBP.