NeRF-Based High-Resolution 3D Environmental Reconstruction Technique Using UAV Video

This paper proposes a method for more efficiently reconstructing high-resolution 3D spaces by combining drone (UAV) and NeRF technologies. Two drone flights were conducted over an area of approximately 2.8 km² targeting a building complex located in Beijing's central business district, collecting approximately 4,000 high-resolution images and utilizing accurate location data to construct reference data. Accuracy, processing time, frames per second, visual quality, and 3D accuracy were evaluated, and RMSE was reduced from 0.023m to 0.017m by considering urban characteristics such as traffic congestion, low light, and high-density architectural environments. In addition, the 3D environment reconstruction time was reduced to 36.5 seconds, achieving real-time performance of 15 FPS. Therefore, this study improved the efficiency of NeRF through flight path optimization and image preprocessing standardization, and demonstrated that more accurate spatial representation is possible in real time in NeRF-based 3D reconstruction. It also proposed future improvement directions utilizing multi-sensor fusion and lightweight technology to overcome the limitations of large-scale and real-time processing.