Multitarget tracking in distributed sensor networks

Abstract

In this paper, we provide a survey of techniques for tracking multiple targets in distributed sensor networks and introduce some recent developments. In the traditional centralized setting, multi-target tracking (MTT) is difficult. There is a combinatorial explosion in the space of possible multiple target trajectories due to the uncertainty in the association of observed measurements with known targets at each timestep. This data association problem has been the primary focus of the MTT literature. Tracking is also complicated by the fact that, for many sensing modalities, targets in close proximity tend to interfere with sensing one another. Compensating for this problem often requires sensing in a higher-dimensional joint space, again increasing computational complexity. Due to the above challenges, MTT is still an open problem even in centralized systems. In distributed sensor networks, we have the additional challenge of mapping an MTT solution onto a sensor network platform with diverse resource limitations, including power, sensing, communication, and computation. Because data collection, processing, and dissemination all come at the cost of resource expenditure, MTT algorithms must make judicious use of resources while simultaneously addressing computational complexity issues. The more recent concepts introduced later in this paper are techniques for addressing these problems by appropriately partitioning the problem into local tasks tracking single targets, which may periodically be combined into small sets of interfering targets. This is combined with other techniques which maintain long-term identity information, explicitly tracking any unresolved confusion between targets and other approaches to resource management based on metrics of the expected usefulness of sensor data for each task.