Multilevel modelling and rendering of architectural scenes

We present a novel approach for multilevel modelling and rendering of architectural scenes using a small set of photographs. Our approach is based on interactive probing of intuitive measures like lengths, widths and heights from single view metrology. These measures can then be aggregated with additional input if need be, for defining high-level primitives such as planes, prismatic blocks, cuboids, spheres, and general surfaces of translation and revolution. The modelling approach is modular and incremental which enables ease and flexibility in building large architectural scenes. Further, our approach allows multilevel modelling and rendering. That is, the model can be enhanced/reduced both in terms of changing details at the primitive level and at the model level. This requires inter-registration of two or more images with common contents. Our approach is computationally simple, fast and robust. We demonstrate our results by building a model of a magnificent historical monument in Delhi - Humayun's tomb.