When interacting with complex three dimentional environments, high demands are put on the hardware in use. There is no doubt, that no matter how powerful the hardware, you can allways find a model complex enough to ruin the performance. It has always been a challange for software developers to, without loss of visual quality, reduce the model in order to lower the amount of calculation that has to be performed.

We have examined different methods for handling complicated models in an efficient way. Starting with methods for optimizing the structure of the model, we went on looking for efficient ways of rendering the geometry.

In the latter case we found an algoritm made by Gernot Shaufler [6] and Jonathan Shade [11] to be of interest. The algorithm replaces parts of the scene with one single textured polygon. The texture is a snapshot of the object from the current point of view. If the parts being replaced is chosen in an intelligent way, the textured quadrilateral (called impostor) can be used in several subsequent frames, resulting in a speedup of rendering time.

To be able to use impostors, the model has to be subdivided into a spacially disjunct hierarchy. For this matter, we used a BSP-tree, optimized for high rendering performance. Thus the BSP-tree alone can be considered a performance improving method.

The algorithm has proven to be able to speed up rendering time significantly. Especially when large models are used. The rendering speedup on the hardware we have used, has reached 50%, but there are reasons to believe that improvements of more than 200% can be achieved. The main reason for not actually getting these results is that the tested hardware does not support some essential operations. The tested graphics hardware that supports these operation has had other flaws. Still, there should exist hardware that combines the features we are looking for. Such hardware would result in higher performance.