Readspeaker menu



Object Tracking

Region based object tracking means tracking an object specified by an initial region through a video sequence. Ferid Bajramovic has implemented a histogram based real-time object tracking system. It is initialized by specifying a region to be tracked, e.g. manually or by using a suitable detector. Subsequently, a region with a histogram most similar to the histogram of the initial region in the first frame is found by local search using the Mean Shift or a Trust Region optimization algorithm.

 

The histogram can be computed from an arbitrary feature, typically color. However, it is also possible to combine the histograms of more than one feature to get a richer representation of the region to be tracked. This extension to the single histogram tracking framework has been published at the DAGM conference 2005 (see below for a list of publications). Details on the algorithms and the extension can be found in Ferid Bajramovic's master's thesis.

 




In the image you can see a tracking example using a video from the CAVIAR project.

View the video in your browser (requires flash player).

Download the video showing the tracking process (slightly better quality).