One of the most desirable skills common to all 3D modellers is the ability to create low polygon models that look good in renderings. If the number of polygons is too high, the results are realistic but it requfires unnecessary system resources and extended rendering times to process the image. Too low, and the rendering will process quickly, but objects may look faceted and computer generated. The goal is to create objects somewhere between these two extremes, and a model submitted by Ernesto Lacalle has demonstrated a very creative way of acheiving these results. The following is the step-by-step process by Ernesto explaining how he created his famous Volkswagon Beetle.
Object Definition and Material Selection
The first step was the selection of object images and materials, which will be required to both model and map the car. We have two goals at this phase of the project. One is to define the shape of the car using scanned images of the object, and the other is to choose images that represent the decals we wish to apply as materials. All the images were scanned from free postcards, brochures and car magazines (see figures 1-4), then edited to remove the unnecessary detail or change the colors.
Creating the Model
The model was made from extruded polylines, which were drawn on raster images of the same scanned views. Only the right half was modeled, then mirrored when completed to assure symmetry of the final model (figure 5). The next step was the conversion to meshes, which was done by exporting the solids to 3ds format then importing them back in again. This process is sensitive to the AutoCAD FACETRES variable, which determines the number of faces when the model is translated from solids to mesh. You can try exporting the model with different values of facetres to find the right one for your use. In this case we needed a very low polygon model, so I decided to loose some detail. In addition, it was not a very accurate model and a low polygon mesh would hide some of the original modeling errors. At this stage the model is converted to polyface mesh. Later I discovered the AccuStudio tools menu which performs this task in an easier way.
The next step was the separation of the original mesh in different sections that could correspond to the different mappings. I did it by exploding the polyface mesh and separating it onto five different layers, one layer for each section that will be mapped with a separate picture. Then I erased the unnecessary faces to make it more efficient and converted each section again to mesh, which welded all the faces in 5 objects corresponding to each of the required maps. We now have individual meshes for the front, rear, sides and wheels (which will require maps) and one object for the bottom and part of the wheels that will have a black material assigned (figure 6).
Then there was the job of determining the right position of every map in the model. Each map had to be applied to each of the 5 sections of the car (figure 7). Once the maps were applied, it was then possible to make the model and the maps even more simple. I was not going to need so many polyface meshes, it could be done with just two. One for the bottom and the other for the rest of the car. We also wouldn't need so many maps, it could be done with just one if we made it semispherical.
The first step was to create the single map. In order to do it, I assigned a self-luminance value of 1 to each of the 5 maps and switched the sun off so that it will be seen from inside. The bottom layer was frozen to prevent blocking any part of the mapped images from the inside, and I chose a panoramic point of view from which all the maps could be seen. The panorama type will be spherical, because we need to cover the top of the car, but as we do not need to map the bottom we choose a very low point of view (Z= 0.00) in order to use half of the entire spherical map, which will eventually be mapped as a dome. Once the image is rendered, it can then be modified in PhotoShop (or any other image editing software) to clean up any seams and imperfections. Multiple versions can also be created by changing the colors and saving each one as a separate file (figure 8).
The spherical mapping is compatible with the spherical rendering since build 245. Before we apply the new map to the model however, we will simplify the model first by welding it into just two separate meshes, one for the bottom and the other for the rest of the car. Then we can assign the spherical decal to the top part of the car taking special care in determining the center point, which has to be the same panoramic point of view from which it was rendered. Once this is done, we now have a very memory efficient model with relativly good detail. You can place hundreds of Beetles in a scene and render without having the problems associated with using detailed 3D models of the car.
Improving the Model
A further step is to add extra realism to our model. There are many improvements that can be done, depending upon your imagination. The one that is shown in the sample image (at the top of the article) incorporated transparent glass windows and side mirror models. You could add more detail, creating some accessories such as the rear antenna, adding an interior or using parts of other models that you may already have.
Here I will describe a method to create the Beetle with real windows that look like glass. A way to do this could be adding an alpha channel. An alpha channel is extra information that some image formats can save within the image as a separate channel. One of these formats is TGA, so save the mapping decal to TGA format. Then add an alpha channel (using PhotoShop or any other image editing software) cutting the transparent part of the windows (figure 9). Additionally we will give the model a silver metallic appearance selecting the entire picture except the red and orange lights, and desaturating it. This will give a silver gray color to the new beetle. You may want to adjust the brightens and contrast to give more life to the map.
We now have to change the decal assigned in AccuRender to recognize the modifications. To do this, under "decal editing" choose the new tga map and check alpha channel using a method so that the glass part will not be seen. Do not check transparent because we do not want the mesh to disappear, we just was to make a hole in the mapping so that the mesh will be seen. We then need to assign a very dark glass material to the mesh (we do not want to be too transparent, otherwise the empty interior will be seen). Finally, in order to give the metallic effect, in the "decal-advanced options" chose Metallic and add a value of reflectivity. We now have a new version of the Beetle suitable for more close-up renderings (figure 10).
Another possibility is a Beetle with all of its lights on. Again this can be done with alpha channels. This night version of our Beetle, ready with all of the lights on can be done in a tga format map, creating an alpha channel for all the lights (it doesn't matter if they are highlights or tail or whatever). Then you should change the decal assigned in AccuRender to the new tga file (the coordinates should remain unchanged) and check alpha channel. Then attach a new decal of the same file (note that the center of the spherical mapping has to be the same, and is in coordinate 0,0,0) and choose again alpha channel, but this time will be inverse (do not check transparent). You will see that one map gives only the car texture and the other only the lights. The last step will be to add a self-luminance value to the map which shows the lights and you have a ready to use beetle for night scenes with all lights on. I am sure that you will think of other new improvements using alpha channels, or any other method.
As you can see, creating the Beetle is a lot of work, and it is only logical in objects that you could use many times. Cars, people and street accessories are the objects I use more of, that's why I can spend the time developing very efficient models that pay for themselves with faster rendering times. Also in this particularly case it was part of an experiment, to develop a modeling method that could be used in new ways.
This test was part of a more ambitious project, which I have not completed yet. Nowadays you can find spherical panoramas on many sites in the web, which could be used to create 3D models for your entourage. There is a great software product called Canoma that does it from flat normal pictures. To use this idea you would need the following:
A job located in a site, which has simple volumes but complex textures
You could take panoramic pictures from certain known points, or have a ready to use panorama taken from a known point.
More or less accurate plans of the entourage
If you end up with an accurate entourage model, and you have the exact coordinates of the panoramic point or points, you could map the entire entourage with a single spherical map projected from that point. This method has an important advantage concerning the efficient use of memory in our system. As the spherical mapping has no fixed dpi resolution but dpg (dots per degree) the final resolution of every object will be proportional to the distance to the object: in other words, the far objects will have low resolution mapping but close objects will have higher resolution, which is perfect for perspective use. The only one condition is that the final point of view shouldn't be far from the original panoramic point, but it doesn't have to be the same. If anyone does it, I will be more than glad to see the results.
rev 8.20.00 :: For more information visit www.accustudio.com