Computational Time-Lapse Video


Department of Computer Science
The University of North Carolina at Chapel Hill
Chapel Hill, NC
Abstract
Sequences of three consecutive time-lapse frames that illustrate the sampling issues of traditional time-lapse methods, and our techniques to reduce aliasing. In a standard uniform sampling, the truck appears in only a single frame, resulting in ``popping''. Adding motion tails makes the truck appear more noticeable and in multiple frames. Using non-uniform sampling chooses additional output frames containing the truck. Finally, we show a result combining both techniques. The tails are shortened because less motion occurs between frames.
We present methods for generating novel time-lapse videos that address the inherent sampling issues that arise with traditional photographic techniques. Starting with video-rate footage as input, our post-process downsamples the source material into a time-lapse video and provides user controls for retaining, removing, and resampling events. We employ two techniques for selecting and combining source frames to form the output. First, we present a non-uniform sampling algorithm, inspired by error-minimizing curve approximation methods, which optimizes the sampling of the input video to match the user's desired duration and visual output characteristics. We present multiple error metrics for this optimization, each resulting in different sampling characteristics. To complement the non-uniform sampling, we present the virtual shutter, a non-linear filtering technique that synthetically extends the exposure time of the time-lapse frames.
People

Eric P. Bennett & Leonard McMillan

Paper

Eric P. Bennett and Leonard McMillan "Computational Time-Lapse Video",
SIGGRAPH 2007 (San Diego, CA)
 Pre-Print PDF

Video

Supplemental Video
This video is encoded with MPEG-4 in Apple QuickTime, available here.

Contact

If you are interested in further details of computational time-lapse techniques, please contact Eric Bennett.

 

 

Visualization of the sampling results, where each vertical line represents a sampled frame. Samplings using the min-error metric choose the majority of their frames from within periods of change and motion to best approximate the video's motion. Alternately, the samplings using the min-change metric avoid frames dissimilar to other frames.