Use of Photogrammetry for Creating Virtual Sets in Video Production

The author explored the use of photogrammetry for creating virtual sets in video production in the diploma thesis. The central aim of the thesis was to determine whether it is possible to build a virtual set using freely available tools and basic photography equipment that would convincingly replace filming at a real location. In the theoretical part, the author defined virtual production as a contemporary production approach, described its historical development and three prevailing approaches, and then presented photogrammetry as a method of three-dimensional reconstruction of the real world. The result was then compared with reference footage of real locations and visual credibility was assessed according to four predetermined criteria. The author found that a high degree of visual credibility can be achieved with freely available tools, but the result depends on the scope and density of capture, not on the properties of the main materials.

Problem

The author explored the use of photogrammetry for creating virtual sets in video production in the diploma thesis. The central aim of the thesis was to determine whether it is possible to build a virtual set using freely available tools and basic photography equipment that would convincingly replace filming at a real location. The problem was identified in the gap between professional practice, which relies on expensive specialised technology and high budgets, and the possibilities of independent production, where resources are limited, equipment is modest, and the team is small or even single-person.

Solution and Method

In the theoretical part, the author defined virtual production as a contemporary production approach, described its historical development and three prevailing approaches, and then presented photogrammetry as a method of three-dimensional reconstruction of the real world. The author examined its use in film and media production, compared software tools available for photogrammetric capture, and justified the choice of tool for the experiment. In the practical part, an experiment was conducted in which two objects were photogrammetrically captured using a digital camera.

From the photographs, three-dimensional models were reconstructed in RealityScan, cleaned and optimised, and assembled into a virtual set in Unreal Engine 5. A person was also filmed in front of a green screen and inserted into the scene using compositing.

Results

The result was then compared with reference footage of real locations and visual credibility was assessed according to four predetermined criteria. The author found that a high degree of visual credibility can be achieved with freely available tools, but the result depends on the scope and density of capture, not on the properties of the main materials. The fully captured object achieved higher scores than the partially captured one.

With the latter, the texture was less sharp because less time and fewer close-up shots were devoted to it. The greatest advantage proved to be flexibility in post-production, as different shots, lighting and camera movements were created from the same model without returning to the location. The approach also proved to be cost-effective, as only free software and equipment already owned were used, but it required a considerable, albeit one-time, investment in learning.