Parallel Online Methods for Lightfield Acquisition

Glasses-free three-dimensional displays, amongst them full-parallax and automultiscopic displays, render samples of the plenoptic function of a scene, which are better known as the light field, to provide a three-dimensional scene impression from multiple viewpoints. The large amount of data that goes together with the sampling process of the plenoptic function demands a sparse representation of the
highly redundant light field data. Within this project, we investigate highly parallelized methods for light field acquisition and reconstruction.

We capture large-scale light field data using our camera rig that consists of 24 cameras with 1.3 MP resolution mounted on a horizontal rail. This rail itself is mounted on a high-precision linear unit, enabling rail movements of up to 2.5m in the vertical and up to 0.3m in the horizontal direction with a precision of 1/300 mm and 1/180 mm respectively. Two control computers that are triggering the cameras
and interfacing the linear unit drive the capturing process.

The special focus of this project is to capture dynamic scenes that contain moving objects. Dynamic scenes demand a proper motion analysis and prediction for the view of inter- and extrapolation in space and time.

The huge amount of data captured by the camera rig contains a lot of redundancies. For example, most of the objects of a scene are Lambertian and
have therefore constant reflection in all directions. In case of dynamic scenes, the principles of standard video coding mechanisms, e.g., sparsity due to static
content, apply as well. Sparse light field representations are supposed to consider spatial as well as time-dependent sparsity for optimal compression.

By using highly parallelized architectures, the ultimate goal is an online compression of the light field data as well as real-time-rendering algorithms
for the light field reconstruction. Currently, we are investigating the sparsity of space-time light fields for optimal representations.

Project Team

Principal Investigators
Prof. Dr.-Ing. Reinhard Koch