Generalised labelled multi-Bernoulli forward-backward smoothing
Michael Beard, Ba-Tuong Vo, Ba‐Ngu Vo · eSpace (Curtin University) · 2016
This paper presents an analytical form for a multi-object smoother, based on a multi-object model known as the generalised labelled multi-Bernoulli (GLMB). The proposed smoother is based on the forward-backward smoothing recursions, which involves a forward pass using the previously developed GLMB filter, followed by backward propagation of a corrector that is used to obtain the smoothed GLMB density. The smoother is derived under the assumptions of the standard multi-object dynamic model, and the standard multi-object measurement likelihood model, i.e. The proposed smoother is capable of handling an unknown and time-varying number of objects, in the presence of measurement origin uncertainty, clutter, and missed detections.