The problem

Dynamic IBL will play a key role in improving the visual fidelity of computer generated imagery for the building design process. High-fidelity rendering techniques, such as image based lighting are physically based and thus require a robust mathematical foundation in order to deliver the desired perceptual equivalence between the real scene and its virtual counterpart. A novel mathematical method for dynamic image based lighting is required to cope with the nature of the video streams now possible with the HDR video camera.

HDR video produces large amounts of data (417 gigabytes (GB) are required to store each second of video), therefore all elements of the processing pipeline must deliver high performance.

The new mathematical method can then be used to define algorithms for IBL using images or videos. Of particular interest are temporal IBL algorithms applicable to the newly acquired HDR video. An additional challenge encountered during the internship was to create interoperability between the PAR and Rhinoceros3D and to extend the PAR so as to run on Arup’s computing cluster. This challenge served as a stimulus to explore other areas of current related research at Warwick Digital Laboratory on high fidelity rendering on shared resources, and was a key element in resolving this issue.

The approach

The mathematical approach to temporal IBL was to extend a spatial sampling technique to the temporal domain. This was achieved by modifying the median cut algorithm into a new scheme called ‘quad cut’. This new sampling scheme is less prone to flickering, although this can still be present. To reduce flickering, temporal filtering of generated samples was applied in an efficient way by splitting and merging samples. These algorithms were developed in Matlab.