Online Monitoring for Multi-Object Measurement Model Parameters

Model-based multi-sensor multi-object tracking approaches crucially depend on their prediction and measurement models. Typically, the approaches work well if the models and their parameters fit the current situation. For a safe operation, it is, therefore, crucial to continuously monitor them. In general, no performance guarantees for the filter can be made in the case of non-fitting models. On the other hand, if the models match, the mathematical properties of the tracking approach, such as Bayes optimality, apply. Distinguishing the two cases leads to a more interpretable and trustable tracking result and provides useful insights for modules later in the processing chain. This paper proposes two methods for monitoring two different parameters of the multi-object measurement model. The evaluation based on simulated data shows that both methods can detect wrong filter parameters, which enhances the reliability of the tracking approach.