In the field of education, virtual reality and augmented reality are significant for realizing online education and creating a multisensorial learning environment. Virtual online education requires effective support for reconstructing 3D representations of real-world objects. In recent years, the Neural Radiance Field (NeRF) is renowned for its high-quality 3D reconstruction capabilities across various fields. However, NeRF still suffers from artifacts named floaters when extended to large-scale scenes such as campuses or museums, which reduces its general applicability in the online education sector. In this paper, we analyze the possible causes of floater generation and reuse the NeRF pipeline to compute scene priors for guiding the geometric reconstruction of large-scale scenes. Experimental results show that our method can effectively remove floating artifacts and provide better background details. Our method achieves competitive performance on metrics compared to baseline models without additional computational overhead.