Multi-object tracking and identity management in wireless sensor networks

The multi-object tracking problem is hard because sensor data has to be associated with specific objects. This is the well known data association problem, which is known to be NP-hard. As a result, all extant algorithms are heuristics and there always exists a finite probability that the tracking system will be confused about the identities of objects after a data association algorithm is applied. We call this issue an identity swapping. To fix the problem, we propose a mathematical framework where the goal is to augment sub-optimal data association algorithms by maintaining a probability distribution over the set of possible identities, thus providing an identity management framework. In identity management, the probability distribution on identities is updated by two events—object mixing caused by objects being in proximity and local evidence incorporation from sensor nodes. Maintaining the full distribution, however, can be computationally infeasible and is often unnecessary—considering that in practice the information provided by this distribution is accessed only in certain stylized ways, such as asking for the identity of a given track, or the track with a given identity. Exploiting this observation, we propose two approximate representations called the marginal belief matrix and the information matrix , and respectively introduce two update operations for both approximations when mixing and local evidence events happen. We analyze and compare the computational complexities of the proposed approximations, and show that these methods provide efficient approximations and exhibit a tradeoff between the two update operations. For their distributed implementation in a wireless sensor network, we propose an agent-based architecture and demonstrate its feasibility by a discrete event-driven simulator we designed for wireless sensor network applications. Based on experimental results from a real-time people tracking system, we conclude that the proposed methods can efficiently fix artifacts of the standard sub-optimal data association algorithms.