LED 버추얼 프로덕션의 데이터 생성 방식 기반 가상공간 제작 유형 분류 및 특성 분석 = Data-Generation-Based Typology and Characteristic Analysis of Virtual Space Creation for LED Virtual Production

다국어 초록 (Multilingual Abstract)

(Background and Purpose) With the enactment of the Virtual Convergence Industry Promotion Act in Korea and the spread of LED Virtual Production (VP), VP has emerged as a core production infrastructure for broadcast, film, and drama. However, in in‑camera VFX environments, a systematic taxonomy and evaluation framework for how Virtual Space should be generated and operated remains underdeveloped. This study aims to systematize the Virtual Space creation methods used in LED VP into four data‑generation‑based categories—manual authoring, camera‑based acquisition, sensor‑based reconstruction, and AI‑based generation—and, from a multi‑criteria decision analysis (MCDA) perspective, to qualitatively compare and analyze the trade‑offs among visual realism, viewpoint freedom, and production efficiency, thereby presenting strategic selection criteria aligned with production goals. (Method) Based on literature review and industry trend analysis, we construct a two‑dimensional framework that cross‑tabulates data generation (creative authoring vs. physical acquisition) and representational structure (limited vs. free viewpoint) and derive a four‑type classification scheme for Virtual Space. Representative use cases and real production projects are then selected, and a qualitative comparative analysis is conducted focusing on indicators of viewpoint degrees of freedom, realism, and production efficiency. (Results) The analysis shows that: ① the manual authoring type offers high 6DoF viewpoint freedom and strong control over narrative and mise‑en‑scene, but exhibits the lowest production efficiency due to high technical complexity and reliance on expert labor; ② the camera‑based type, which employs live‑action plates and 360° HDRI, achieves high optical realism and operational efficiency, but its viewpoint freedom is limited by parallax constraints stemming from the absence of depth information; ③ the sensor‑based type, using LiDAR, photogrammetry, and NeRF/3DGS reconstruction, provides the highest geometric accuracy and scale consistency, but entails substantial post‑processing workload and rights‑management requirements; and ④ the AI‑based type is efficient in terms of turnaround time and cost through prompt‑based synthesis, yet faces constraints in securing physical consistency and visual stability depending on training data and licensing conditions. (Conclusions) None of the four types simultaneously maximizes realism, viewpoint freedom, and production efficiency, confirming the need to design hybrid pipelines that combine multiple types according to project genre, budget, and studio configuration. By systematizing LED VP Virtual Space creation methods in terms of data‑generation methods and representational structures and proposing an MCDA‑oriented evaluation framework, this study provides a theoretical basis for examining the balance between physical sets and Virtual Space, workflow design, and investment efficiency at the VP planning stage. Future work will apply type‑specific Virtual Space samples to real studio shoots and conduct quantitative and qualitative evaluations with directors, cinematographers, and technical directors regarding perceived realism, immersion, and ease of pipeline integration, in order to empirically validate the proposed taxonomy in real VP environments.