Producing photorealistic landscapes on a 3D city model platform

Abstract

An interactive 3D city model visualization system using high resolution IKONOS imagery and the corresponding Rational Polynomial Coefficients (RPC) camera model has been developed at CRISP. This paper describes further improvement to the photorealistic quality of our visualization system with an additional module that supports rendering of vegetation features, such as trees and shrubs, to compliment the existing terrain and building features. The challenges of implementing the additional module are: (1) to draw a huge number of trees with minimal Central Processing Unit (CPU) overheads, so as to reserve the CPU for more computationally intensive calculations for critical operations by the system; (2) the view management of each individual tree. To retrieve the trees, the vegetation types on the high resolution geo-referenced IKONOS image are classified into a few classes according to the shape and diameter of the canopy. The canopy with a circular or star shape, and diameter more than 2.5 meters, is extracted as an individual tree. The trees in dense or sparse areas are randomly located and different classes of the trees are issued with specific plan-view and side-view billboard images generated from a commercial software suite. To implement the tree module into our platform, we efficiently draw multiple instances of the tree billboard image using hardware accelerated techniques, incurring minimal CPU overheads. We allocate the first vertex buffer to contain the fixed set of geometry for a single tree and subsequently allocate the second vertex buffer for the position, orientation and color for all the trees. This reduces the amount of data which is needed to be supplied to the renderer, hence dramatically improving performance. View management is implemented at the Graphics Processing Unit (GPU) level, using High Level Shader Language (HLSL) to handle the angular calculations, rotations and image selection for each tree.