Methods to model the motion of extended objects in multi-object Bayes filters

The multi-object Bayes filter represents all objects in the environment using random finite sets. Due to the set representation, a filter state represents all objects in the environment. Thus, object interactions can easily be integrated. Since an integration of object interactions due to their extent into the prediction step requires the determination of arbitrary motion models for each object, a subsequent validation of the predicted multi-object state is proposed. The validation is possible by minimizing the weight of invalid predicted states. An alternative approach is the usage of hard-core point processes in order to thin the particle sets. This contribution presents an integration of three different validation methods into the multi-object Bayes filter: weight adaption, thinning, and a hybrid method using thinning and weight adaption. The performances of the proposed validation methods are compared using simulated and real world sensor data.