Multi-sensor multi-object joint detection and tracking from image observations using labeled multi-Bernoulli densities

In this paper, we consider a multi-object joint detection and tracking problem in the framework of Bayesian filtering of image data collected from multiple sensors. Our method is based on the random finite set (RFS) representation of multi-object states and recently developed labeled multi-Bernoulli (LMB) filter. Under the assumption that multiple sensors are independent conditional on the multi-object state, closed form of multi-object state update procedure is derived using the probability generating functional (p.g.fl.) methods. Exact parameter update of the posterior multi-object densities is discussed as well. And combined with an inherent labeling scheme, the proposed method is able to form multiple tracks of individual moving targets. The multi-object posterior density can be updated iteratively, regardless of order, using the image data captured in each sensor at each sampling instance. Simulation results show that the proposed solution can successfully detect and track multiple targets in typical surveillance scenarios.