AI-Driven Target Localization Enabled by UAVs and IoT

Abstract

Environmental monitoring plays a crucial role in various applications, including disaster response, infrastructure management, and industrial operations. One of the key challenges in environmental monitoring is source localization, which involves identifying and pinpointing specific targets or sources within a designated Area of Interest (AoI). This thesis addresses the problem of multiple source localization (MSL) in scenarios involving Internet of Things (IoT) sensors and Unmanned Aerial Vehicles (UAVs). Existing solutions to the MSL problem suffer from inevitable limitations, such as increased algorithmic complexity as the number of sources grows, primarily due to their inherent iterative nature. Moreover, these solutions tend to exhibit inaccurate localization of multiple sources in scenarios with sparse sensory data, necessitating a large number of sensors and, thus, an increased localization cost. The research introduces a novel approach that leverages conditional generative adversarial networks (cGANs) and persistent homology-based peak-finding with sub-pixel refinement to formulate the MSL problem as separate image-to-image translation and peak-finding sub-problems. Extensive experiments, both on real-life and synthetic datasets, demonstrate the proposed approach’s efficacy compared to state-of-the-art models. Results show significant improvements in mean localization error, false alarm rate, missed detection rate, and optimal sub-pattern assignment error distance, even under conditions of sparse sensor placements. In a related study, the thesis extends the MSL problem to scenarios involving UAVs. The approach combines cGANs with adaptive sampling techniques based on Bayesian Optimization and introduces a novel acquisition function to guide UAV data collection. The research demonstrates the effectiveness of this approach through rigorous evaluation against similar methods, showcasing substantial enhancements in MSL accuracy. Overall, this thesis presents innovative solutions to the challenges of MSL in environmental monitoring, offering robust and resilient algorithms that can adapt to sparse sensor data and achieve accurate source localization in both IoT and UAV-assisted scenarios.

Date of Award	14 Dec 2023
Original language	American English
Supervisor	Shakti Singh (Supervisor)