Approaches To Localization And Tracking In Wireless Sensor Networks

Due to the requirements in civilian and military applications, wireless sensor net-works (WSN’s), especially those for surveillance purposes, have drawn considerable attention in last three decades. In surveillance wireless sensor networks, target local-ization and tracking are two main problems. Localization is often performed by using the information of time-of-arrival (TOA), time-difference-of-arrival (TDOA), acoustic energy or received signal strength (RSS). The localization problems based on this in-formation are typically highly nonlinear and non-convex, which are difficult to solve. The traditional method applies the first-order Taylor-series expansion to linearize the nonlinear problems. However, its performance cannot satisfy the requirements in large noise environments. Target tracking problems are very difficult to deal with due to the nonlinearity of the state and measurement models. Current nonlinear target tracking algorithms either have poor performance, or have high computational complexity, in-cluding the Extended Kalman Filter (EKF), the Unscented Kalman Filter (UKF), the Particle Filter (PF), and the Gaussian Mixture Model (GMM) algorithm. In addition, of all target tracking problems, maneuvering target tracking is one of the most challenging problems. In the multiple-model algorithms that deal with maneuvering target track-ing problems, using an improper transition probability matrix (TPM) may degrade their performance. To overcome these drawbacks, we employ convex optimization technique to solve the localization problems based on the TDOA, acoustic energy, and RSS mea-surements, the tracking problems based on the range-only and TDOA measurements, and the TPM estimation problem in maneuvering target tracking. Specifically, the main contributions of this thesis are shown as follows:1. In the TDOA-based localization problem, we resort to the Monte Carlo impor-tance sampling (MCIS) technique to find an approximate global solution to the non-convex ML formulation. This method has low computational complexity and its performance approaches the Cramer-Rao Bound (CRB) even at high noise levels.2. We propose an approximate weighted least squares (WLS) formulation for acous-tic energy-based localization, and perform the semidefinite relaxation technique to approximately solve this problem. Simulation results show that the SDP method performs better than several existing methods at high noise levels. 3. For noncooperative and cooperative sensor network localization problems, we propose a new WLS formulation using RSS measurements. After solving the WLS problem, its solution is used as a starting point to solve the original ML problem through a local search algorithm. This method has lower computational complexity than the existing semidefinite relaxation method.4. For range-only and TDOA-based tracking problems, we respectively study an angle-parameterized maximum a posteriori (APMAP) method and an approxi-mate maximum a posteriori (MAP) method, both of which have good perfor-mance and low computational complexity.5. In Jump Markov Systems (JMS’s), we give a low-complexity ML estimation method to estimate the TPM based on the greedy strategy.