Distributed target tracking over a low-cost sensor network.

Proliferation of low-cost, lightweight, and power efficient sensors and advances in networked systems enable the employment of multiple sensors. Distributed estimation provides a scalable and fault-robust fusion framework with a peer-to-peer communication architecture. This thesis investigates the problem of distributed target(s) tracking over a low-cost sensor network and proposes several efficient and reliable algorithms to address the aforementioned problem. The primary issue of using low-cost sensors in distributed target tracking is how to balance between communication cost and convergence performance. To overcome this difficulty, we develop a new sample greedy gossip distributed Kalman filter for distributed single-target tracking over a low-cost sensor network. The proposed algorithm leverages the information weighted fusion concept and a new sample greedy gossip process. Instead of finding the optimal path of each node in a greedy manner, the proposed approach utilises a suboptimal communication path by performing greedy selection among randomly selected active sensor nodes. Theoretical convergence analysis and uniform boundedness are also performed to support the proposed algorithm. The main feature of the new algorithm is that it provides fast convergence rate with relatively low communication overload. In distributed multi-target tracking, each sensor node runs a local multi-target tracking filter for multi-target state and identity estimation. The issue of current multi-target tracking algorithms is that they usually suffer from exponential complexity and require prior knowledge on the environment, e.g., target detection probability and clutter rate. This hinders the application of low-cost sensors in multi-target tracking as these sensors usually have limited computational capability and offline calibration is also not cost-effective. To address these sensor-related practical issues, we propose a polynomial-time joint probabilistic data association filter using stochastic Gibbs sampling technique and incorporate it with a multi-Bernoulli filter to accommodate the unknown environmental parameters. It is theoretically proved that the proposed solution provides a performance-guaranteed approximation and is demonstrated to show promising performance in a dynamic environment. We then consider the problem of accurate Gaussian mixture approximation for implementing joint probabilistic data association filter when targets are moving closely and propose a novel information-driven approach to tackle this problem. The proposed information-driven joint probabilistic data association algorithm is obtained from the minimisation of a weighted Kullback-Leibler divergence to approximate the posterior Gaussian mixture probability density function. The resulting approximation approach turns out to show similar structure as the generalised covariance intersection in sensor fusion. Theoretical analysis reveals that the proposed approach with ideal detection probability guarantees boundedness of the error covariance and yields unbiased estimation. Different from single-target tracking, local multi-target estimations contain data association uncertainty and therefore this issue requires careful adjustment in sensor fusion. Through an equivalent information form, we extend conventional fusion strategies, i.e., consensus on measurement and consensus on information, to multi-target tracking scenarios using joint probabilistic data association filter. This means that data association uncertainty and sensor fusion problems are solved simultaneously. Motivated by the complementary characteristics of these two different fusion approaches, a novel distributed multi-target tracking algorithm using a hybrid fusion strategy, e.g., a mix between consensus on measurement and consensus on information, is proposed. A distributed counting algorithm is incorporated into the tracker to provide the knowledge of the total number of sensor nodes. The new algorithm developed shows advantages in preserving boundedness of local estimates, guaranteeing global convergence to the optimal centralised version and being implemented without requiring global information, compared with other fusion approaches. To support the implementation of distributed multi-target tracking, we finally investigate the problem of practical implementation of multi-dimensional assignment. To address this problem, we propose two efficient implementation algorithms using Tabu search and Gibbs sampling. As the rst step, we formulate the problem of generating the best global hypothesis in multi-dimensional assignment as the problem of finding a maximum weighted independent set of a weighted undirected graph. Then, the meta-heuristic Tabu search with two basic movements is designed to find the global optimal solution of the problem formulated. To improve the computational efficiency, we also develop a sampling based algorithm using Gibbs sampling. The problem formulated for the Tabu search based algorithm is reformulated as a max product problem to enable implementation of Gibbs sampling. The detailed algorithm is then designed and the convergence is also theoretically analysed. The performance of the two algorithms proposed are verified through nu-merical simulations and compared with that of a mainstream Lagrangian relaxation implementation algorithm.