Fast-Forward Video Visualization and Adaptive Fast-Forward Techniques for Video Surveillance: Institut für Visualisierung und Interaktive Systeme

Information for

Logo VIS

Fast-Forward Video Visualization and Adaptive Fast-Forward Techniques for Video Surveillance

Analysing surveillance video footage often includes watching the video in fast-forward mode.

Playback speed adaption is a technique for comfortable viewing surveillance video footage. According to a provided relevance measure the playback speed is adjusted to equalize the cognitive load. Hence, periods of less activity are shown in fast-forward, while in periods with much activity playback speed is reduced. This way, users are able to recognize important details in scenes with much activity, while boredom is reduced for periods without change. This is a clear benefit over conventional fast-forward at constant pace, as the following image illustrates:

 

 

Additionally, the question arises how the accelerated videos should be shown best. Here, the video visualization should allow for good object identification as well as good motion perception. Moreover, adapting the video playback speed according to its content requires to communicate the current playback velocity.

This page includes the supplementary material for the paper "Learning a Visual Attention Model for Adaptive Fast-Forward in Video Surveillance" and "Information-Based Adaptive Fast-Forward for Visual Surveillance", which are methods for adapting the playback speed in video, and "Evaluation of Fast-Forward Video Visualization", which introduces and evaluates four video fast-forward visualizations and three adaptive fast-forward playback speed visualizations.

Publications

2012

Evaluation of Fast-Forward Video Visualization
Höferlin, Markus; Kurzhals, Kuno; Höferlin, Benjamin; Heidemann, Gunther; Weiskopf, Daniel: Evaluation of Fast-Forward Video Visualization. In: IEEE Transactions on Visualization and Computer Graphics (TVCG): No. 12 (2012), pp. 2095-2103.
[XPS] [PDF] [DOI] [OpenXML] [BibTeX] [Slides] [Details]
Learning a Visual Attention Model for Adaptive Fast-Forward in Video Surveillance
Höferlin, Benjamin; Pflüger, Hermann; Höferlin, Markus; Heidemann, Gunther; Weiskopf, Daniel: Learning a Visual Attention Model for Adaptive Fast-Forward in Video Surveillance. In: In Proceedings of International Conference on Pattern Recognition Applications and Methods (ICPRAM), pp. 25-32, 2012.
[XPS] [PDF] [DOI] [OpenXML] [BibTeX] [Slides] [Details]
total number: 2

2011

Information-Based Adaptive Fast-Forward for Visual Surveillance
Höferlin, Benjamin; Höferlin, Markus; Weiskopf, Daniel; Heidemann, Gunther: Information-Based Adaptive Fast-Forward for Visual Surveillance. In: Multimedia Tools and Applications: No. 1 (2011), pp. 127-150.
[XPS] [PDF] [DOI] [OpenXML] [BibTeX] [Slides] [Details]
total number: 1

Evaluation of Fast-Forward Video Visualization

Download Video (41 MB)

Learning a Visual Attention Model for Adaptive Fast-Forward in Video Surveillance

Example 1


Example 2

Comparison between conventional fast-forward, and adaptive fast-forward based on motion activity, rényi divergence, and visual attention model on the Crowded Airport Sequence. The conventional approach accelerates the video to a constant playback speed and discards the frames in between, while the adaptive approaches dynamically accelerate the sequence according to their relevance measure. The video sequence is averagely accelerated by factor 3.


Comparison between conventional fast-forward, and adaptive fast-forward based on motion activity, rényi divergence, and visual attention model on the Airport Sequence. The conventional approach accelerates the video to a constant playback speed and discards the frames in between, while the adaptive approaches dynamically accelerate the sequence according to their relevance measure. The video sequence is averagely accelerated by factor 5.




Download video (14.5M)
Download video (34.6M)


Example 3


Example 4

Comparison between conventional fast-forward, and adaptive fast-forward based on motion activity, rényi divergence, and visual attention model on the Noisy Airport Sequence. The conventional approach accelerates the video to a constant playback speed and discards the frames in between, while the adaptive approaches dynamically accelerate the sequence according to their relevance measure. The video sequence is averagely accelerated by factor 5.


Comparison between conventional fast-forward, and adaptive fast-forward based on motion activity, rényi divergence, and visual attention model on the Night Sequence. The conventional approach accelerates the video to a constant playback speed and discards the frames in between, while the adaptive approaches dynamically accelerate the sequence according to their relevance measure. The video sequence is averagely accelerated by factor 10.



Download video (34.6M)
Download video (51.8M)



The video files are encoded with the H.264 codec and provided in a .mp4 container. If you have any problems viewing the video sequences, you may download the VLC media player, including several useful codecs: VLC


Information-Based Adaptive Fast-Forward for Visual Surveillance

Example 1

Comparison to constant fast-forward

Example 2

Performance comparison under noise

Comparison of standard fast-forward and our approach on the Airport Sequence. The conventional approach accelerates the video to a constant playback speed and discards the frames in between. Our approach dynamically accelerates the sequence according to the information gain and blends frames based on the physiological model. The video sequence is averagely accelerated by factor 5.


Comparison of two different adaptive fast-forward approaches on the Noisy Airport Sequence. The first is depending on Horn-Schunck motion magnitude, the second is the proposed information-theoretic approach. In contrast to motion measure, the proposed information-based approach is robust to noise and leads to good results. The video sequence was averagely accelerated by factor 10.


Video 1
Video 2

Example 3

Comparison of visualization techniques

Comparison of standard video fast-forward visualization (discard frames) and our physiologically motivated blending approach on the Airport Sequence. The video sequence is averagely accelerated by factor 7.

Video 3




The video files are encoded with the H.264 codec and provided in a Quicktime .mov container (Quicktime 7 or above). If you have any problems viewing the video sequences, you may download the VLC media player, including several useful codecs: VLC