Tracking, track-to-track fusion and surveillance

Date of Completion

January 2010


Engineering, Electronics and Electrical




In this dissertation, three problems in the area of tracking, track-to-track fusion and surveillance are investigated. The first one is the problem of tracking (state estimation) in the presence of severe measurement nonlinearity known as “the contact lens problem” (curved uncertainty regions), which appears in very long range tracking scenarios using, e.g., a phased array radar. It is shown that existing nonlinear filtering approaches have consistency problems or have significant loss in range accuracy. In this dissertation, a novel filtering approach — the measurement covariance adaptive extended Kalman filter (MCAEKF) — is presented. For the very long range tracking problem considered, the filter is shown to have superior consistency and tracking accuracy. ^ Second, for distributed multisensor tracking systems, the problems of Track-to-Track Fusion (T2TF) and Association (T2TA) are investigated. First, under the assumption of synchronicity, the exact algorithms are presented for T2TF without memory (T2TFwoM) and with memory (T2TFwM), for the information configurations of fusion with no, partial and full information feedback. The impact of information feedback on the fusion accuracy is also investigated. It is shown that information feedback has a negative impact on the fusion accuracy of T2TFwoM, and, in contrast, information feedback is beneficial in T2TFwM. Later, exact and approximate algorithms for the asynchronous T2TF are also developed for different information configurations. It turns out that the AT2TF algorithms based on the generalized Information Matrix Fusion (IMF), even though heuristic, are remarkably robust. They show good consistency over the practical range of process noise levels and have close to optimal fusion accuracy. Due to the simplicity of their implementation, the algorithms are appealing candidates for practical applications. Then, for the problem of T2TA, the exact sliding window test, which uses track estimates within a time window, is derived. Counterintuitively, it is shown that the belief “the longer the window, the greater the test power” is not necessarily correct. ^ Finally, a real-time cooperative path decision algorithm for surveillance using a group of Unmanned Aerial Vehicles (UAV) is proposed. To handle the multiple competing objectives in the surveillance, a layered decision framework is proposed, where different objectives are deemed relevant at different decision layers according to their priorities. An important objective of the path decision algorithm is to navigate the UAV safely in the hostile environment. To achieve this, it is shown to be necessary to increase the time horizon of the path decisions. Accordingly, a Rollout Policy is proposed which has moderate complexity and, when used in the layered decision framework, it is shown to have superior performance compared to the conventional combined objective one-step lookahead path decision approach. ^