NeRFOrtho: Orthographic Projection Images Generation based on Neural Radiance Fields

The application value of orthographic projection images is substantial, especially in the field of remote sensing for True Digital Orthophoto Map (TDOM) generation. Existing methods for orthographic projection image generation primarily involve geometric correction or explicit projection of photogrammetric mesh models. However, the former suffers from projection differences and stitching lines, while the latter is plagued by poor model quality and high costs. This paper presents NeRFOrtho, a new method for generating orthographic projection images from neural radiance fields at arbitrary angles. By constructing Neural Radiance Fields from multi-view images with known viewpoints and positions, the projection method is altered to render orthographic projection images on a plane where projection rays are parallel to each other. In comparison to existing orthographic projection image generation methods, this approach produces orthographic projection images devoid of projection differences and distortions, while offering superior texture details and higher precision. We also show the applicative potential of the method when rendering TDOM and the texture of building façade.