MOTEXATION: Multi-Object Tracking with the Expectation-Maximization Algorithm

The paper proposes a new edge-based multi-object tracking framework, MOTEXATION, which deals with tracking multiple objects with occlusions using the Expectation-Maximization (EM) algorithm and a novel edge-based appearance model. In the edge-based appearance model, an object is modelled by a mixture of a non-parametric contour model and a non-parametric edge model using kernel density estimation. Visual tracking is formulated as a Bayesian incomplete data problem, where measurements in an image are associated with a generative model which is a mixture of mixture models including object models and a clutter model and unobservable associations of measurements to densities in the generative model are regarded as missing data. A likelihood for tracking multiple objects jointly with an exclusion principle is presented, in which it is assumed that one measurement can only be generated from one density and one density can generate multiple measurements. Based on the formulation, a new probabilistic framework of multi-object tracking with the EM algorithm (MOTEXATION) is presented. Experimental results in challenging sequences demonstrate the robust performance of the proposed method.