direkt zum Inhalt springen

direkt zum Hauptnavigationsmenü

Sie sind hier

TU Berlin

Inhalt des Dokuments

MA: Anomaly Detection through Video Generation


 The CIT research group created over the last years several different anomaly detection algorithms. One technique is called IFTM and uses behavior learning to model the normal behavior of a given datastream. Based on the trained model, values of the datastream can be forecasted and compared with the current monitored values. Thus, high differences are reported as anomalies based on a dynamically adapting threshold function.

Based on this method, an anomaly detection has to be created for videos. As advances in video generation and forecasting through deeplearning techniques happened over the last years, there exist first possibilities to integrate such techniques into IFTM.

Check for example the work by: http://www.cs.columbia.edu/~vondrick/tinyvideo/

In this thesis, the following tasks are included:

  • Select public available evaluation datasets
  • Include suitable video generation techniques into the IFTM framework
  • Create visualizations for the algorithmic results (videos, static images and helpful diagrams)
  • Evaluating the results for anomaly detection

Prerequisites for working on this topic are advanced knowledge in algorithmic design and good programming skills in either Java or Python.



Zusatzinformationen / Extras

Quick Access:

Schnellnavigation zur Seite über Nummerneingabe

Auxiliary Functions


Florian Schmidt
+49 30 314 28306
Room TEL 1205