© 2000 - 2011 LIN - Leibniz Institut für Neurobiologie Magdeburg

LIN: Forschungsabteilungen > Akkustik, Lernen, Sprache > Unterpunkt Ebene 3 > Unterpunkt Ebene 4

Titel: LIN Layout

Home switch to german Print Search:
Facebook YouTube
Staff Links Sitemap


 Objecttracking in colored image sequences

System model

Figure 1: System model

Already for some years we are working on the image sequence analysis. We are doing this in a close co-operation with the IESK of the University of Magdeburg. The specific interest is thereby on the use and application of neurobiologically motivated neuron models and structures.

As the result of our work a system was developed, which is distinguished by the combination of classical image processing methods and neural approaches. This architecture contains both the feature filters for the texture analysis and a neural network for the storage of these features in form of an associative memory.

The objects determined in the figure-ground-separation are normalized by a translation process (figure object completion). After a training phase the MHA based associative memory is then able to store and recognize arbitrary objects and their instances (prototypes). This characteristic can be used for object tracking in the image scene. Due to the used translation procedure and the cluster characteristics of the neural net, translation and rotation-invariant object recognition is possible.

Results from object tracking

Figure 2: By means of the associative characteristics of the MHA it is possible to track objects with image disturbances (left picture and upper picture, red characterized) in an image scene (right picture shows the learned prototype of the object).

last update: 2010-02-24 report a bug print this page