As part of my development of the proposition I developed a set of localised objectives, primarily as a means of ensuring this project explored aspects of camera projection where questions remained that had not been addressed sufficiently by the first project or indeed had emerged from my work on that project. I found it helpful to utilise these as a framework for reflecting on the second project.
Taking the time to consider and articulate the workflow at the outset proved to be highly beneficial as a means of identifying all the key stages, the software pipeline, and also specific instances where ‘round-tripping’ between software applications would occur. This was heavily influenced by my decision to adopt a new approach to creating the artwork (Jones 2010), which necessitated the development of pre-defined geometry and an animated shot camera at the very beginning of the process. I was unfamiliar with this approach prior to my research but found it offered the potential to address a number of variables that inevitably arise on shots with more expansive camera movement. However, the limitation in this approach was that it didn’t address the critical process by which the projection cameras are defined and tested before any meaningful artwork is undertaken and I believe my notion of the ‘projection strategy’ goes some way to addressing this deficit. The tool I developed to facilitate and simplify this process was thoroughly tested and offers proof of concept. Implementation of this workflow did necessitate development of the projection geometry to a higher level of detail than I had used previously and conflicted with the arguments made following my work on project 1, that this should be as simple as possible. The additional level of detail was entirely for the purpose of generating self-shadowing information from the ambient occlusion pass and, whilst I could have developed the geometry with much less detail and simply painted in the shadows using conventional painting techniques, I found this to be a quicker and more accurate approach. The level of detail applied to the projection geometry can clearly be adjusted to facilitate different workflows. Moreover, I encountered an instance where additional sculpting work was needed to project the contours along the edge building 3 façades which suggests that the level of detail should be considered in isolation for every project and, in some instances, for each projection. My workflow (Figure 24) presents as sequential and linear, and it would therefore be logical to assume that this translated to execution of the project. However, in reality the project revealed unexpected results at almost every iterative step and requiring some ‘back and forth’ between the various stages of the workflow to resolve the particular issue.
This project presented instances where the texture area provided by a specific projection was insufficient at various points along the shot camera’s motion path resulting in gaps in coverage. Several methods for addressing the problems were demonstrated, the most elegant being those that extended the field of view of the projection camera, either by Overscanning or by breaking the camera from the motion path and backing it off. In both cases I proposed the use of non-destructive approaches and was able to utilise a projection rig, constructed for this purpose, to configure and test the projections before committing to matte painting.
It is possible to discuss texture smearing and doubling in unison, despite these manifesting differently and having different causational factors. I found prevention to be the primary means of resolving these issues and my projection strategy was invaluable in this regard. The checkerboard rig system presented both artefacts with great visual clarity allowing different configurations for the projections to be tested. I found that smearing could be significantly reduced simply by finding the best position and angle for projecting the texture. However, areas where there was extreme changes in perspective were more problematic but could be overcome by distributing the whole projection over several cameras and limiting the projection of each to a smaller surface area, using alpha channels to define the boundaries. My approach to the ground surface was the most obvious example (Figure 44).
Texture doubling was largely prevented by identifying geometric objects that overlapped other objects, from the perspective of the shot camera, at any point in the motion path and simply projecting these separately. However, this approach was only possible where each geometric object was separate and could not prevent textures doubling onto faces within the same object. In these cases, the only viable solution was to patch the offending area by projecting an area of texture, again defined by an alpha channel, and laying it over. Indeed, patch projection was used to fill small areas of coverage and smearing in addition to doubling, most of which were identified during the diagnostic stage using the projection rig, but others were discovered later in the process when the sections of matte painting had been added and replaced their respective checkerboard texture.
Maintaining consistent resolution between all the projections became an issue where I had extended the camera’s field of view, either by Overscanning or by backing it off from the motion path of the shot camera. I adapted the tool developed for testing the projections so it would enlarge the checkerboard pattern and thereby provide a clear visual clue when differences were arising between adjoining or overlapping projections. resolution was being lost. My tool allowed the resolution to be raised until the scale of the checkerboards aligned and therefore allowed this compensation to be made and factored into the matte painting. The methodology and application provided proof of concept, although this must be mitigated against the relatively small changes in resolution in this project. Further investigation is needed to establish its effectiveness where there are extreme changes of resolution, perhaps an implementation of a deep descent into an environment utilising the “Powers of Ten” model, presented by the designer Charles Eames’s in the 1970 Norton lectures at Harvard and continues to be employed to catalyse the building of more accurate conceptions of scale. (Jones et al., 2006, p.194).
Peter Holmes PhD Journal 