Disney Demos Automated Multicamera Editing
Algorithm determines center of attention and calculates cuts
August 12, 2014
VANCOUVER, B.C., CANADA
—Disney Researchers have come up with a way to auto-edit together footage
from multiple social cameras. The team created an algorithm that takes in
multiple video feeds and outputs a “single video of user-specified length that
cuts between the multiple feeds automatically,” according to their
Researchers Ido Arev, Hyun Soo Park, Yaser Sheikh, Jessica Hodgins and
Ariel Shamir, said their algorithm uses “existing cinematic rules.”
By determining the center of attention of the cameras and, by default, their
operators, the algorithm indicates where the cuts should go. The center of
attention is key. It is derived from the orientation of each person involved in
a group activity and the footage they capture. The aggregate of the resulting
viewing angles is referred to as the “gaze concurrence” or “3D joint attention,” which produces a metric for the “spatial and temporal location of the important
‘content’ of the activity.”
This information is used to plot a path on a trellis graph to determine edit
cuts while avoiding jump cuts and crossing the line of action. The algorithm is said to cut according to the
action and automatically switches between levels of zoom.
“While this indicator does not describe a complex narrative flow as a human
editor might…. It does provide the information needed to create a watchable
cutoff the video that condenses the video footage [a] hundredfold or more while
retaining the important action that occurred during the event,” they said.
comments powered by Disqus.