A Fast Labeled Multi-Bernoulli Filter based on Joint Likelihood for Multiple Extended Target Tracking

Multiple extended target tracking can provide the positions, shapes and trajectories of targets simultaneously due to high-resolution measurements. However, the states of the irregular extended targets are difficult to estimate efficiently and accurately, this paper proposesa fast labeled multi-Bernoulli (LMB) filter based on a joint likelihood function. The filter simultaneously estimates target number and trajectories, including kinematic and extension states. Gaussian components are updated by the measurement transformation filter, modeling targets as star-convex set. A joint likelihood function fuses target dynamics and shape properties to measure similarity between targets and measurement cells. To reduce complexity, extended targets are grouped by segmenting an undirected graph, and data association within groups during joint prediction/update is solved using Gibbs sampling. Simulations show the filter improves both tracking performance and computational efficiency for multiple extended targets of varying shapes.