Inference over distribution of class probability for probabilistic object classification

State of the art Bayesian classification approaches typically maintain a posterior distribution over possible classes given available sensor observations (images). Yet, while these approaches fuse all classifier output thus far, they do not provide any indication regarding how reliable the posterior classification is. On the other hand, current deep learning based classifiers provide an uncertainty measure, thereby quantifying model uncertainty, however, do so on a single frame basis and do not consider a Bayesian sequential framework. In this paper we develop a novel approach that infers a distribution over posterior class probabilities within a Bayesian framework, while accounting for model uncertainty. This distribution enables reasoning about uncertainty in the posterior classification, and thus is of prime importance for robust classification and object-level perception in uncertain and ambiguous scenarios, and for safe autonomy in general. The distribution of the posterior class probability has no known analytical solution, thus we approximate this distribution using two methods: sub-sampling and Unscented Transformation. We evaluate our approach in simulation and using real images fed into a CNN classifier.