Five-Dimensional
(5-D) Depth-Velocity Filtering for
Enhancing
Moving Objects in Light Field Videos
Experimental
Results from an upcoming paper by
Chamira U. S. Edussooriya, Don G. Dansereau, Len T. Bruton and Panajotis Agathoklis
Five-dimensional (5-D) light field video (LFV) (also known as plenoptic video) is a more powerful form of representing information of dynamic scenes compared to conventional three-dimensional (3-D) video. Enhancing objects moving with constant velocity at constant depths as well as exposing occluded regions of a scene can be accomplished by applying 5D depth-velocity filters on LFV. The approach is based on analyzing the region of support (ROS) of the 5-D spectrum of such objects and designing depth-velocity filters which encompass in their passbands the ROS of the moving objects. The performance of 5-D depth-velocity digital filters (proposed in the upcoming paper) in enhancing moving objects in LFVs is illustrated below using a LFV generated using the commercially available Lytro light-field (LF) camera.
Central sub-aperture of the original light field Video (LFV)
The Original LFV is comprised of a blue car (object of interest), and a red and an orange trucks (interfering objects). All three were manually moved in planes approximately parallel to the camera plane and a commercially available Lytro LF camera was employed to capture static LFs of the scene. The distances between the camera plane and the planes corresponding to motion of the blue car, red truck and the orange truck are 162 cm, 158 cm and 110 cm, respectively. The blue car and orange truck were manually moved from left to right with the speed of 1 cm per frame whereas the red truck was manually moved from right to left with the speed of 1 cm per frame. Note that the red truck moves at approximately the same depth of the blue car so that the red truck is very difficult to be attenuated by using 4-D depth filters.
Central sub-aperture of the 5D depth-velocity filtered LFV
It can be observed in the 5D depth-velocity filtered LFV that the object of interest (blue car) is enhanced with slight distortion while the two interfering objects (red and orange trucks) are substantially attenuated. In the 30th frame the blue car (object of interest), which is completely occluded in the original LFV, is considerably exposed in the filtered LFV. Further, the noise present in the original LFV is significantly attenuated by the 5-D depth-velocity filtered.