STRL is developing an algorithm to estimate the appropriate shooting position for a scene composed of multiple 3D models, to make video production more efficient. This technology automatically finds the best camera position to shoot images focusing on a specified 3D model.
Estimating the appropriate camera position to shoot images with best subjective qualities
STRL conducted experiments to collect a large amount of subjective evaluation data about the perceived quality of images shot from different positions. Based on the relationship between this evaluation data and the features of each object, STRL developed an algorithm to estimate the appropriate camera position to shoot the best images.
Applicable for scenes containing multiple 3D models
Given multiple 3D models in a scene, the 3D model in the foreground sometimes occludes the other 3D models in the background, making it difficult to predict the appropriate shooting position with the existing technique. The proposed technique takes into account the effect of each 3D model on the goodness of views for the whole scene, making it possible to predict the appropriate shooting position for a scene composed of multiple 3D models.
STRL will work on improving the processing speed to support animated scenes by around 2025. We will also apply this technique to adaptively select proper expressions depending on different viewing styles by around 2030.
- Akihiro Sakai, Yasuhito Sawahata, Yamato Miyashita, and Kazuteru Komine. 2022. Cognition-aware automatic viewpoint selection in scenes with crowds of objects. In SIGGRAPH Asia 2022 Posters (SA '22). Association for Computing Machinery, New York, NY, USA, Article 21, 1–2.
(You will leave the NHK website)