I have numbered the buildings for ease of reference during the discussion on this page.
Building 1
Checkerboard texture projected from a duplicate of the shot camera taken on frame 1. Full coverage is seemingly achieved and no stretching or doubling of textures is evident
Timeline advanced to frame 30. Note how areas of missiong coverage appear on the far left and the upper section of the façade. The left side is covered by street furniture so may not be a problem.
Frame 27 of the shot camera identified as the optimum frame for accessing all the missing coverage and a plausible patch solution. However the angle makes placement of the alpha to blend the two projections difficult. Also, because the front façade would comprise of two images, problems would arise later in the pipeline; say for appending the façade with light, shadows, or decals such as signage, as they would have to transcend two different image sources.
Timeline advanced to the final frame in the motion path. Missing coverage is now visible in the door recess and also the side façade. Some textures doubling through.
Timeline on frame 30 but this time field of view extended by changing focal length of projection camera from 24 (default) to 17. Full coverage is achieved and patch projection no longer necessary on this façade.
Also tried breaking from the camera path and backing off the projection camera, which yielded a similar result.
Slightly larger checkerboard squares indicate loss of resolution.
Patch applied to the side façade using a duplicate of the shot camera taken from the final frame of motion path.
Temporary alpha applied here using a Roto node but refinements later in Photoshop may allow patch to be extended to include the loss of coverage in the door recess.
I found that I could use the same projection camera to achieve coverage on building 2. Moreover, because these two building do not overlap at any point in the motion path, there is no risk of textures doubling from one building to the next.
Building 3
Initial projection of the primary texture using a camera derived from the final frame of the shot camera but broken off the path and backed up slightly to maximise coverage in the extreme foreground.
However, the perspective at which the camera displays this façade begins from a more orthographic perspective and becomes more sheer as the shot progresses. Consequently this projection smears across onto the far side of the building, evident in the stretching of the squares in the checkerboard pattern.
My second iteration included a second projection camera used to patch the entire left side of the façade.
A Roto node is used to add a temporary alpha on the patch to delineate the boundary between the two textures at the point of the protruding brick pillar. The intention is to create the final alpha channels in Photoshop but this is a useful temporary solution for projection testing, acknowledging it lacks the accuracy needed for a final solution.
There is overlap between building 3 and building 2, only partial at the outset but almost full overlap by the end of the sequence.
Texture from the primary projection camera on building 3 is doubling onto the side façade on building 2.
Both buildings are combined within the same projection
This provides an example of where textures doubling from overlapping geometry can be prevented by using a separate projection setup for each building. However this can only work if the geometry are separate objects and would not resolve texture doubling between different surfaces on the same object.
The projection camera required to patch this side of the façade needs to be more directly facing the façade to prevent texture smearing and cannot be achieved from a location close to the motion path of the shot camera. However, from this position there is a mismatch in the resolution of both projections on this façade.
The focal length calculation could be used to derive an accurate resolution for the patch projection except, rather than the previous application to achieve coverage, this approach involved a comparative analysis on the size of the squares on the checkerboard textures applied to both the patch and the primary projection and adjusting the resolution of the patch until it matched the primary projection visually.
Ground Projection
Projected from a camera derived from the motion path at frame 65, broken out, raised slightly on Y axis and angled down to give a more orthographic view of the street.
From this frame the resolution looks OK and the checkerboard squares are maintaining their shape.
Seen from frame 1 of the shot camera, stretching of the texture is evident in the foreground due to the extreme change in perspective. Also a large area of coverage is missing, which could possibly be patched but there is no clear point at which the alpha could be drawn to blend the patch. Preferred option would therefore be to devise a camera from somewhere close to frame 1 and use this to project this entire section of the ground. Alpha could be drawn along the kerb edge.
Seen from frame 1 of the shot camera, Projection of foreground (red) surface from a camera derived from the shot camera at frame 1 but broken off, raised slightly on Y axis and angled down to give a more orthographic view. No loss of resolution or significant texture stretching is evident.
At frame 30 areas of texture doubling are evident on the front and side sections of the road. This can be patched.
For the right side of the road, the centre projection does not provide sufficient detail of the kerb edges that would be visible in the early frames.
Seen from frame 1 of the shot camera, these surfaces (blue) are projected from a camera broken from the motion path and moved significantly to the left, creating a more orthographic perspective.
The checkerboard was applied a little clumsily and therefore already appears stretched.
However when seen from frame 50, the pattern has not changed, which suggests that the position of this projection camera is OK.
No loss of resolution stretching is evident.
The following is an audio-visual discussion of development of the ground projection
Street Decals & Furniture
Steps & Pillar
Projection of the steps and pillar using a copy of the camera used to apply the primary textures to building1 and 2. Both elements are projected separately to prevent doubling.
Both elements only feature for a short time, fully out of shot by frame 35. In this period there is no discernible texture smearing and therefore no requirement for revision of the position or field of view of the camera or patch projection.
Bollards & Streetlamps
As a second iteration, I added the bollards and streetlamps to the same projection.
This holds together until around frame 70 when some loss of coverage is apparent.
A second projection for the street items with the camera derived from the last frame of the motion path but broken off and adjusted slightly to give more of a ‘side-on’ view. This allows the offending areas of missing coverage to be patched.
A second version of each DMP was created and some light information painted in.
The intention will be to blend these together in the composite and use some form of random on the opacity of the ‘lit’ version to present a flicker effect, a common characteristic of first generation electric street lighting.
Peter Holmes PhD Journal 