Real Time Rendering Techniques

Modeling the castle

First and foremost to begin the project a suitable scene was designed using Auto Cad and modeled using the 3DS Max software package. When modeling any scene it is essential that everything is to scale and each element is proportional to another. This was the reasoning behind using Auto Cad to model the scene as it an architectural program by which designs can be made to scale. This model provided the plans which were then used to model the scene in 3DS Max. In order to do this two sets of drawings were made, the first was a top down Birdseye view of the model, secondly an elevation plan which described the structures height.

Matsumoto Castle

Many of the castles that were researched whilst planning the project were considered however most castles all look rather bleak, old and diminished. Therefore it was decided that rather than your traditional European castle, a castle from Japan named Matsumoto was planned.

A vast period of time was then spend modeling the castle in 3ds max, as a highly detailed and accurate model of the castle was developed. Once the model was completed, another extended period of time was spend creating the texture maps for the model. One cravat to creating the model was discovered when a first attempt to load the model into direct X is that as the model had multiple hi-resolution textures the 3DS loading module unfortunately did not support the material mapping functionality that was implemented in 3DS MAX. There were two ways around this problem.

Firstly, each individual component of the castle could be separated, a new texture map could be generated and loaded in as separate objects into DirectX. The other method would be to write a new 3DS object loader that fully supported material mapping functionality. Due to time constraints, rather than re-inventing the wheel, the objects of the castle were separated into individual components and loaded into DirectX. One problem that occurred at the beginning was due to the highly detailed models, the application took around 30 – 60 seconds to load in all of the meshes. Therefore in order to address this issue, many of the complex and highly detailed mesh data was excluded and the model was compressed into around 222,000 vertices.

In order to texture the model each mesh was un-wrapped and a texture was individually hand made for each one which was a time consuming process. Bump and normal maps were applied to the Rocky and wooden features of the model to provide additional detail. All of the textures were then pre-baked to include bump mapping and lighting calculations. This again took a very long time to process.