Adapting the state uncertainties of tracks to environmental constraints

In most multi-target tracking algorithms it is assumed that the movements of the targets are statistically independent of each other. This assumption may lead to predictions which are not possible due to physical exclusions. Instead of integrating the dependence between the objects directly into the tracking module, we propose to handle scenarios with interactions between the tracked object and other objects by adapting the uncertainty about the state of the object. The adaption is based on occupancy grids and reduces the uncertainty without endangering the consistency of the tracking filter.