Home /Projects /Hybrid Rendering

 

 

Fig 1 - fac.jpg

Hybrid Rendering

Photorealistic and geometrically accurate environmental visualisations can be used in simulators, engineering project overviews, computer games and cinematic media. Hybrid Rendering is concerned with creating environments with these characteristics from limited sets of photographs and laser scans. The work ties in with the SceneStealer work being done at EdVEC, extending the simple geometric models produced by SceneStealer to include extensive surface detail derived from a laser scanner. SceneStealer allows photogrammetric reconstruction of a simple geometric model of an environment from a limited set of photographs of the environment. The photographs are then projected back onto the model using view-dependent texture mapping.

 

Fig 2 - skin.jpg

The Hybrid Rendering system takes as inputs a SceneStealer environment (Figure 'fac.jpg') plus laser scan data from the same environment (Figure 'skin.jpg'). The system improves the quality of the SceneStealer model greatly by registering the laser scan data with it. An approximate registration is computed from at least four point-correspondences between the simple model and the laser data and then this approximation is used as an initial guess for aligning the data sets accurately using the ICP algorithm. Concave regions of fine detail in the laser data may be obscured after registration by lying 'behind' SceneStealer polygons. A hole-cutting facility in the system enables these hidden details to be revealed (Figure 'plastic.jpg'). After registration the data can be exported into a suitable format for rendering as high quality ray-traced images and video ('churchHoles2.mpg'). The system can export the laser data as a point cloud of tiny spheres or as a polyhedral skin and can interpolate in texture mapping using special frequency-dependent interpolation(link to Watson, O'Brien & Wright 2002 paper) surface shaders.
 
Watch mpeg movie - churchHoles2.mpg (1MB)
 

Fig 3 - plastic.jpg

 
_