Research on drones positioning system based on visual navigation

Abstract

This thesis investigates the design and implementation of a drone positioning system based on visual navigation techniques. The objective is to enable accurate and robust localization of drones by matching aerial images with corresponding satellite views, addressing the challenges of viewpoint variations, scale changes, and occlusions. The proposed system adopts a cross-view geo-localization framework that utilizes deep learning models to extract discriminative features from multi-perspective imagery. A ConvNeXt-based backbone network is employed to enhance feature representation capabilities. Additionally, a multiple-classifier structure is introduced to capture rich contextual information and improve robustness against position shifts and scale variations. The system was trained and evaluated on publicly available drone and satellite image datasets. Experimental results demonstrate that the proposed method achieves superior performance compared to previous approaches in terms of localization accuracy and robustness. This research contributes to advancing drone autonomy and navigation reliability in real-world environments.