Chapter 4: Mixtures of Dynamic Textures

Motion Segmentation Results

These are examples of motion segmentation using the mixture of dynamic textures. The original video is on the left, and the segmented video is on the right. The video is in Quicktime format (H.264). A comparison with other motion segmentation methods is also available.

Segmentation of Synthetic Video

The first four synthetic examples are from Doretto, et. al., "Dynamic Texture Segmentation", in ICCV 2003. These video are segmented using an initial contour provided in the paper.


[mov (0.3 MB)]

Synthetic video where two patches of the ocean has been rotated 90 degrees.


[mov (0.3 MB)]

Synthetic video where the dynamics of two patches of the ocean have been altered.


[mov (0.8 MB)]

Synthetic video of two textures: ocean and fire. The segmentation follows the changing outline of the flame.


[mov (1.3 MB)]

Synthetic video of two textures: ocean and steam.

Segmentation of Synthetic Texture Database

The synthetic texture database contains 299 videos of synthetic textures with 2, 3, or 4 segments.

synthdb2 - video textures with 2 segments. (view all)

[mov (0.4 MB)]

[mov (0.4 MB)]

[mov (0.3 MB)]

[mov (0.6 MB)]
synthdb3 - video textures with 3 segments. (view all)

[mov (0.6 MB)]

[mov (0.5 MB)]

[mov (0.2 MB)]

[mov (0.3 MB)]
synthdb4 - video textures with 4 segments. (view all)

[mov (0.2 MB)]

[mov (0.6 MB)]

[mov (0.5 MB)]

[mov (0.5 MB)]

Segmentation of Real Video

These are examples of segmenting real video using the mixture of dynamic textures.

water fountain

[mov (0.8 MB)]

Fountain with three types of water motion.

highway traffic

[mov (0.2 MB)]

Motion segmentation of a traffic highway scene.

bridge traffic

[mov (0.2 MB)]

Motion segmentation of traffic on a bridge.

pedestrian scene

[mov (1.2 MB)]

Pedestrians moving in different directions on a walkway.

crowded pedestrian scene

[mov (2.0 MB)]

Crowded pedestrian scene. Lots of people moving!

Hour-long Pedestrian Segmentation

This is a segmentation of an hour-long pedestrian video using the mixture model learned from the "crowded pedestrian scene". Note that this segmentation required no reinitialization at any point, or any other type of manual supervision. The sequences contain a fair variability of traffic density, various outlying events (e.g. bicyclies, skateboarders, or small vehicles, pedestrians changing direction, etc.) and variable environmental conditions (e.g. varying clouds and shadows). The video is sped up by 2 times, and each clip is about 5 minutes long.

[mov (24 MB)]

[mov (17 MB)]

[mov (13 MB)]

[mov (11 MB)]

[mov (14 MB)]

[mov (9.5 MB)]

Clustering Example

An example of clustering 253 traffic videos into five clusters. Six typical sequences from each cluster are shown in the five rows, which perceptually correspond to light traffic (spanning 2 clusters), medium traffic, slow traffic, and stopped traffic ("traffic jam").

mov (0.5 MB)]

Copyright Antoni Bert Chan 2008