Ground Terrain

This ground terrain is arguably the single most important layer in this project. This is because the image transcends from the immediate foreground all the way through the mid-ground to the distant hills.

On all matte paintings, which depict a space that recedes from the camera, these elements are notoriously difficult to get right in the 2.5D environment. This is because the image is 2D and therefore holds only horizontal (X) and vertical (Y) properties. The computer therefore expects the image to be presented by assigning to geometry on a vertical axis.

However, once the image is assigned, it needs to present tangible depth (Z axis) so that other elements align correctly to it. Of particular note are the various shadows from all the separate elements that have been painted into the ground layer (Fig 1). This would require application of the texture to geometry that is placed on the ground, rather than standing upright.

Fig 1: Ground image.

When we use a vertically orientated card (Fig 2) by assigning the texture to its UV’s, the image looks correct in 2D view.

Fig 2: Ground image applied to a vertically aligned card in 2D view.

However in 3D view we begin to see the issue (fig 3). The card sits in a single position on the Z axis that is in front of the sky and distant mountains but its position is arbitrary in relation to the foreground and mid-ground areas that the image is depicting visually.

Fig 3: Ground image applied to a vertically aligned card in 3D view.

So the issue then becomes where do we place other scene elements relative to the card holding the ground texture?

Figure 4 shows a test render, which was recorded with the ground terrain applied to a vertically orientated card and with the gate and foreground tree element temporarily placed in a plausible (front left) area of the scene.

One problem is the the rear/uppermost part of the ground layer riding up and contaminating the distant mountains. However the real giveaway is how the foreground elements appear to float and slide across the surface of the ground terrain.

Fig 4: Test render with ground image applied to a vertically aligned card.

All other elements that need to be placed in either the foreground or mid-ground will be similarly affected because they all need to reside in Z space relative to the ground. It is therefore not difficult to conclude that this particualr method of texture application is unsuitable for this section of the matte painting.

The clear implication is that the card, rather than being orientated vertically, needs to be rotated 90° on its X axis so it lays flat to the ground (Fig 5).

Although still far from perfect, there is a marked improvement in the stability of the ground elements over the course of the shot. Moreover, a further benefit to from this approach, is that elements within the texture (such as contact shadows) now occupy specific positions in depth (Z Space) and can therefore be used as depth cues for aligning other (separately projected) elements.

Fig 5: Test render with ground image applied to a horizontally aligned card in 3D view.

Applying the Ground Terrain to the Card UV’s

The card is laid flat and the image applied to the card UV’s (Fig 6). The texture looks OK in 3D view, although the gap between the ground and distant hills suggests that the card need to be pushed back in Z.

Fig 6: Ground image applied to a horizontally orientated card in 3D view

In 2D view (Fig 7) the image is clearly very different to the original matte painting

Fig 7: Ground image applied to a horizontally orientated card in 2D view

We can scale and transform the Card but it is almost impossible to replicate the original ground section of the matte painting. This is because the image is applied to the UV’s of the geometry but, because this is laid flat, and therefore at a glancing angle, to the shot camera, there are infinite potential permutations across the transform and optical attributes of the geometry.

Projecting the Ground Image on to a Card

Projecting the texture onto the geometry from the ‘hero’ frame, appears to provide a much more plausible solution. Figure 8 shows the setup in 3D view.

Critical to this method are the characteristics of the projection camera. In this case I moved to a position along the timeline where there matte painting was at its most visible, which in this case was the very last frame in the sequence. I then made a duplicate of the shot camera on that frame so the position and rotation values were retained. I then removed all the keyframe data so the duplicate camera effectively because a static instance of the shot camera at its final resting position. I made no changes to the optical attributes of the camera. This became my ‘projection camera’.

Fig 8: Ground texture projected onto horizontally orientated card geometry in 3D view

In 2D view we can see that, because the image is applied from the same position as the shot camera, the projection creates a perfect replica of this section of the matte painting (Fig 9)

Fig 9: Projected ground texture in 2D view

Dealing with the Ground Undulations

In the ground image there is clear undulation in the terrain (Fig 10a) but, if the image is applied to just a Card node then these undulations will not be present in 3D (Fig 10b).

Fig 10a: Ground image projected onto Card, viewed in 2D through shot camera	Fig 10b: Card on which ground image is projected

The issue is again one of alignment. Imagine the hay bales on the right side of the mid-ground, which, in the matte painting, are clearly located at various heights on the ground undulations (Fig 11).

Fig 11: location of the hay bales in the matte painting

When, in due course, we come to align these bales to their respective shadow on the ground plane, some will be levitating above the ground plane and the gap between the bale and the ground will result in visible sliding during the camera move.

This can be overcome by adding some height to the ground geometry so it aligns to the contours in the 2D image and there are two ways this can be addressed without resorting to a 3D modelling application. One would be to use deformation attributes built into the Card node and another would be to add additional geometry (i.e. spheres) and project the image onto multiple geometric objects.

Bicubic Card Deformation

The Card deformation potentially allows the entire terrain to be created in a single node. However this stretches the UV’s and therefore introduces a risk of creases or hard edges forming that would be visible when the image is projected. This method is also more difficult to modify when aligning to the visual contours in the image.

Figs 12a and 12b show application of cubic deformation. This uses X and Y subdivision to generate curve handles that can be pulled in order to generate the deformation. In this case 8 subdivisions were applied in each axis.

Fig 12a: Cubic deformation on Ground Card in 3D view	Fig 12b: Cubic deformation on Ground Card in 2D view

Fig 12c shows how the hill is created by raising the curve handle. However what is evident here is that there are insufficient handles to create the different tiers of undulation evident in the image.

By increasing the subdivisions, we generate more control handles but still insufficient to allows these tiers to be created successfully.

Fig 12c: sub-divisions raised to 9 x 9

Only by raising the subdivisions to 24 x 24 (Fig 12d), do we get close to having enough control over the deformation site. However the demands on the computer resources are such that the software takes an eternity to refresh so successful manipulation is highly problematic.

Fig 12d: sub-divisions raised to 24 x 24

Additional Geometry

Figure 13 shows the ground image projected onto the Card and spheres. Curved shapes such as spheres or horizontally aligned cylinders can look quite crude when applied in this way, akin to Tellytubby Land. However, once the image is projected, these can be manipulated quickly and easily using simple transforms. By using several overlapping instances of a shape, the complexity of the terrain can be emulated.

Fig 13: spheres added and transformed

By adding these geometric elements we begin to create visible undulations on the ground surface in 3D (Fig 14). Large areas of the spherical geometry intersects the card and lies underneath. However, because this is never seen by the shot camera, it can be ignored.

Fig 14: Undulations apparent in 3D

Fig 15 shows the ground in 2D mode. From here it is easy to align the shapes using numeric value changes in the respective node properties, whilst viewing the effect of these changes in 2D. The huge advantage here is being able to see the scene from the shot camera whilst changing 3D attributes and with immediacy of feedback in the viewer.

Fig 15: 3D undulations can be finessed using 2D view

Overall the use of multiple geometric shapes as a means of creating the undulations in the terrain is quicker, much easier to fine tune and less demanding on computer resources.