Interface and method of interfacing between a parametric modelling unit and a polygon based rendering system

Abstract

An interface for use in a 3-d graphics system comprising a parametric modelling unit for modelling objects as high order surfaces, and a polygon based rendering system for rendering polygon modelled objects for display. The interface comprises an input for receiving data and a subdivision unit coupled to the input for processing the data. The interface includes a converter coupled to the subdivision unit for determining from leaf patch data a first plurality of values representing vertices of tessellating polygons describing the leaf patch, and for determining from sub-leaf patch data a second plurality of values representing the vertices of tessellating polygons describing the sub-leaf patch. The interface also has a combiner, coupled to the converter, for combining the values to form leaf polygon data defining the polygon vertices at a first subdivision level, and an output coupled to the combiner for outputting the leaf polygon data.

Description

[0001] This invention relates to an interface for use in a 3-d graphics system comprising a parametric modelling unit and a polygon based rendering system.BACKGROUND TO THE INVENTION[0002] Traditional 3D rendering systems use a mesh of polygons (usually triangles) to model the objects within a scene. Triangles have the advantage that they are of a simple form, and it is therefore relatively easy to perform operations such as transformation, lighting, texturing and rasterization on them in order to produce an image for display.[0003] A known alternative to modelling objects using a mesh of polygons is to segment the object into areas and fit a number of curved, high order surfaces, frequently termed patches, to the different areas of the object being modelled. These patches are generally defined as parametric surfaces with the surface shape governed by a grid of control points. An advantage of using high order surfaces is that the set of control points usually requires a much smaller...

AI Technical Summary

Problems solved by technology

A major disadvantage of modelling using high order surfaces is that it introduces added complexity in the rasterization stage of the rendering system.

Rational patches have certain modelling advantages over the simpler non-rational variety but can be slightly more complex to process in some respects.

We have appreciated that for a hardware implementation this is not practical for two reasons: internal (i.e. on-chip) storage is expensive and the retrieval of large amounts of externally stored data can cause a bottleneck in the performance of the system.

A second problem which occurs when interfacing between parametric data and polygon based data in a combined graphics system is that different levels of subdivision may be required to convert parametric data relating to a first patch and parametric data relating to a second patch, where the first and second patches represent adjacent areas of the object being modelled.

If conversion of adjacent patches is not constrained to apply the same level of subdivision to each patch, then cracks can appear in the modelled object because the surface with the higher level of subdivision has extra sample points and thus potentially a slightly different shape.

However, this results in some patches being excessively subdivided when they could be tessellated adequately at a less divided level and does not target the processing to the areas where it is necessary to produce adequate results.

Unfortunately, such connecting meshes may introduce abrupt changes in surface direction.

We have also appreciated that stitching the edges of adjacent patches together is computationally intensive and requires the stitching mesh to be regenerated as the subdivision level changes.

Although this scheme is mathematically correct, it has a subtle flaw.

Computer graphics hardware has limited precision and is unable to represent exact polygon vertex locations.

It is therefore frequently impossible to exactly place a flattened vertex on the shared boundary.

This leads to problems caused by "T-Joints", as well known in the art.

We have appreciated that these T-Joints could be fixed by again introducing stitching polygons,.however this is a rather inefficient approach since such polygons are tiny and the `set-up` costs involved in polygon rendering would be better utilised on more significantly sized polygons.

This methord generates vast amounts of data and is not really suitable for real-time rendering.

1) Subdivide all patches of the model to the same subdivision level, This is simple but wasteful of resources.

It is not ideal as the stitching polygons are additional costs and can induce abrupt changes in direction (as seen in FIG. 5a).

3) Use Clark's method, allowing non-uniform subdivision even within patches, but suffer from T-Joint problems.

4) Apply further stitching polygons to Clark's method.

5) Subdivide the mesh into micro-polygons.

A further problem with known interfacing techniques is so-called polygon popping.

Thus increasing the subdivision ratio even a small amount may cause the actual level of subdivision to increase dramatically.

Unless this process is well controlled it causes undesirable visual effects in the animation as shown in FIG. 3.

Unfortunately, this calculation can frequently fail.

This means that one or both of the first partial derivates may be zero, resulting in an incorrect normal.

This is rather unfortunate as such surface arrangements of control points are often needed to model completely valid shapes and can easily be generated by modelling software packages.

This will usually provide an adequate result, but is costly in terms of additional calculation.

Unfortunately, even this approach can fail.

Images