Multi-target tracking and localization in distributed wireless sensor networks

Tracking and localization are enabling techniques for many applications using wireless sensor networks, robotic networks and ad-hoc networks. Classic positioning approaches are mostly based on time-of-arrival (TOA) information only and direction-of-arrival (DOA) information only, which require collecting and processing, hence consume considerable bandwidth for information transmission. This dissertation takes an alternative approach of joint TOA/DOA processing which allows a single sensor node to localize the targets to reduce the cooperation and communications among sensor nodes. Accordingly, several key technical issues are investigated in this thesis for tracking and localization with data fusion in the context of distributed wireless sensor networks. These issues are: Cramer-Rao bound (CRB) analysis of location estimation without and with data fusion, sensor selection for optimal data fusion, sensor management framework for multi-target tracking in static wireless sensor networks, joint optimization of mobility control and sensor allocation for sensors in mobile sensor networks. All these issues are motivated and treated under the considerations of limited network resources (e.g. power, bandwidth and computing) of wireless sensor networks, which require the strong need for distributed sensor management.;Proper sensor management hinges on the understanding of network performance, which is assessed analytically in this dissertation by deriving the CRB of target location estimation under two types of data fusion: measurement fusion and state fusion. The analysis reveals what factors would affect the accuracy of measurements and location estimation, and further quantifies how senor-target geometry would affect the positioning accuracy. Based on these CRB-related metrics, optimization problems are formulated for sensor selection which leads to the best target localization accuracy given the sensing resources when optimal data fusion is adopted.;For multi-target tracking in wireless sensor networks, sensor management is an essential task in order to balance the tracking performance and costs under limited network resources. There are two important questions during sensor management: 1) how are a group of sensor nodes dynamically allocated to track targets in order to optimize the future tracking performance, and collaborate within the group via data fusion, 2) how to implement the management and tracking scheme in a distributed fashion at low computational complexity and scalable network cost. For mobile sensor networks, controlled movement of sensors must be also considered in accordance with the sensor allocation to optimize the tracking quality. By building on the CRB analysis results for location estimation, this dissertation develops a multi-layer architecture for multi-target tracking in static and mobile sensor networks. Constrained optimization problems based on the information theoretical perspective are formulated to maximize the expected overall tracking performance for all targets. Through local one-hop communication and limited information exchange with neighbor nodes, each node autonomously decides on whether to participate in tracking tasks. The optimizations are relaxed to the convex problems for computational feasibility, and implemented in a distributed manner using iterative subgradient search to achieve global optimality.