Traffic Classification and Retrieval
These are video from Chapter 3. Videos are in Quicktime format (h.264).
Motion Model
The motion is modeled using the dynamic texture, a generative probabilistic model.
The following example shows the model of the motion flow of light, medium, and heavy traffic.
The original sequences contains 50 frames and the model is extrapolated to 100 frames.
[mov (1.1 MB)]
Classification Examples
This is an example of classification of traffic congestion under sunny, overcast, and nighttime lighting conditions.
The classifier was trained only with daytime sequences, yet still performs well on nighttime scenes.
Below are a few test sequences with their classification results.
[mov (0.7 MB)]
Retrieval Examples
This is an example of the retrieval system. The far left column
contains the query video, and the video to the right are the top five returns. The system performs equally well with query videos taken during the daytime or nighttime. Some of the video have raindrops on the camera lens, which cause
occlusion and blurring (3rd and 5th results of the 2nd night query).
[mov (0.7 MB)]
[mov (0.7 MB)]
Copyright Antoni Bert Chan 2008