A light-field is defined by all light rays traversing a specific area. To record all light beams within a scene, researchers from the moving picture department of the Audio and Media Technologies division use a robot system that captures slightly different perspectives of a scene. Information such as the depth of the scene objects can be extracted. These depth maps are then used as a basis for various effects and post-editing steps like camera panning and refocusing – all with a single light-field recording.
Especially when it comes to immersive media content acquisition for VR and mixed reality applications, experts from Fraunhofer IIS consider light-field technology a promising approach. An essential benefit of light-field data is the possibility to generate different perspectives from any position of a scene. Ultimately, this will allow a navigation with more than 3 degree of freedom (DOF) with limited additional transitional movement, also known as 3DOF+, which enables looking behind objects and persons.
For their paper “Acquisition system for dense lightfield of large scenes“ the scientists captured natural dense light-field images and offer them for free download. In contrast to existing data-sets that have been obtained using hand-held light-field cameras the published data offers a much higher variety in perspective. Like this, the data-set allows for comparing image- and depth-image based rendering (IBR, DIBR), testing depth estimation algorithms, comparing virtual views to a reference and many more. The overall size of the scene and the camera-system is very comparable to typical media production environments. For more information on the light-field data, take a look here.