A Real-Time Orthophoto Generation Approach for UAV Based on Deep Learning Visual Feature

In recent years, there has been a growing demand for uncrewed aerial vehicles (UAVs) to generate real-time orthophotos across various fields, including emergency response, surveying, and land assessment. Deep learning for visual feature extraction has proven highly effective, showing robust performance in complex environments characterized by dynamic scenes, low-texture areas, and significant appearance variations. In this study, we propose a real-time UAV orthophoto generation approach based on deep learning. Our approach uses simultaneous localization and mapping (SLAM) pipeline based on deep features to estimate camera poses and georeferencing. Orthorectified images are seamlessly mosaicked in real-time using optimal orthogonality values and multiband blending algorithms. Experimental results indicate that deep features contribute to accurate orthophoto generation. Compared with current state-of-the-art methods, our approach produces high-quality orthophotos at a reduced computational cost.