For a profound understanding of traffic situations including a prediction of traf-
fic participants’ future motion, behaviors and routes it is crucial to incorporate all
available environmental observations. The presence of sensor noise and depen-
dency uncertainties, the variety of available sensor data, the complexity of large
traffic scenes and the large number of different estimation tasks with diverging
requirements require a general method that gives a robust foundation for the de-
velopment of estimation applications.
In this work, a general description language, called Object-Oriented Factor Graph
Modeling Language (OOFGML), is proposed, that unifies formulation of esti-
mation tasks from the application-oriented problem description via the choice
of variable and probability distribution representation through to the inference
method definition in implementation. The different language properties are dis-
cussed theoretically using abstract examples.
The derivation of explicit application examples is shown for the automated driv-
ing domain. A domain-specific ontology is defined which forms the basis for
four exemplary applications covering the broad spectrum of estimation tasks in
this domain: Basic temporal filtering, ego vehicle localization using advanced
interpretations of perceived objects, road layout perception utilizing inter-object
dependencies and finally highly integrated route, behavior and motion estima-
tion to predict traffic participant’s future actions. All applications are evaluated
as proof of concept and provide an example of how their class of estimation tasks
can be represented using the proposed language. The language serves as a com-
mon basis and opens a new field for further research towards holistic solutions
for automated driving.