Fraunhofer-Institut für Optronik, Systemtechnik und Bildauswertung IOSB

Hybrid AI Combining Machine Learning and Dynamic Bayesian Networks for Behavior Recognition

Bachelor- / Master Thesis in cooperation with Airbus Defense and Space GmbH


Modern monitoring networks are able to provide trajectories of all kind of vessels and aircrafts within worldwide or at least ex-tended environment. Best known are Automatic Dependent Surveillance – Broadcast (ADS-B) and Automatic Identification Sys-tem (AIS) used within air and maritime surveillance. It is foreseeable that ongoing trends like Internet of Things (IoT), digitalisati-on, automotive, smart cities and decentralisation (blockchain) will enable additional systems in the near future – e.g. in traffic control and monitoring. The real challenge becomes the related situational awareness and the estimation of the intent of the tracked objects. Activity recognition and the determination of suspicious situations are a significant challenge. To address these topics in the context of trajectories several technics are available. Here, two techniques should be considered:

  • Deep Learning and
  • Dynamic Bayesian Networks (DBN)

Deep Learning demonstrates to deliver fascinating results. However, large data sets are needed. This makes it difficult to apply Deep Learning if these data sets are not available. Further, Deep Learning techniques lack to be explainable, but this is im-portant to estimate confidence in the AI results and also to have a chance to certificate the techniques for critical applications. On the other hand, there are model-based methods like Bayesian Networks. These methods are explainable due to their very nature. They depend on expert knowledge during their design. New studies apply these DBNs more and more to complex envi-ronments – especially for situation assessment and anomaly detection, e.g. the detection of illegal, unreported and unregulated (IUU) fishing.  The focus of this thesis is on the combination of DBNs and data-driven methods, like deep learning, for trajectory based anomaly detection:

  • Modelling and implementation of a Dynamic Bayesian Network for anomaly detection of a maritime or airborne incident (multi object scenarios).
  • Combination of AI Machine Learning with DBN to a hybrid method, i.e. the generation of evidences for the DBN based on AI Machine Learning models.


Field of Study

  • (Economic) Computer Science, Engineering Sciences, Mathematics


  • High interest in the field of artificial intelligence (AI)

  • Knowledge in the field of machine learning / deep learning

  • Pleasure in contributing your own ideas and good communication skills

We offer

  • Opportunity to apply your knowledge and skills in a scientific environment

  • Open and communicative working atmosphere and intensive supervision

  • Working on trend-setting topics in the field of artificial intelligence

  • Supervision of seminar papers and final papers

  • High practical relevance through cooperation with the company Airbus Defense and Space GmbH