Robust Incremental Bearing-Only Localizer
Almozahmi, Mohammed Hamad Mothana
Almozahmi, Mohammed Hamad Mothana
Citations
Altmetric:
Supervisor
Department
Robotics
Embargo End Date
Type
Thesis
Date
2026
License
Language
English
Collections
Research Projects
Organizational Units
Journal Issue
Abstract
Drowning remains a major cause of accidental death worldwide, and the time window for a successful intervention is often only a few seconds or minutes. In open-water settings (beaches, lakes, and rivers), (i) to be able to detect that a person is in distress and (ii) to be able to localize a person fast enough to guide an. Autonomous rescue asset. Many existing methods rely on human monitoring or the victim carrying a device, and these assumptions do not hold in many unregulated or crowded environments. This thesis presents a vision-based maritime rescue framework that localizes a person in distress using only bearing measurements from three unmanned surface vehicles (USVs). Each observer USV has been outfitted with a monocular camera and estimates the direction of the target in its local image frame. Using USV heading, these angles are transformed into global bearings and fused to estimate target position on the sea surface. Subsequently, the estimated value is dispatched to direct a rescue USV that can manoeuvre to the victim and deploy a flotation device. To get a statistically consistent estimator under realistic measurement noise, we formulate a scalar line-constraint measurement model based on the perpendicular residual to each bearing ray. We furthermore derive a distance-aware variance model that captures how small angle errors are expanded into meter-level lateral uncertainty with range. Building on these models, we The lightweight estimator, called RIBOL (Robust Incremental Bearing-Only Localizer), combines bearing constraints with a constant-velocity motion prior and performs outlier rejection through gating. We validate the approach in two stages. First, planar Monte Carlo simulations quantify localization accuracy is affected by factors such as bearing noise, formation geometry, and measurement dropouts/outliers; secondly, a vision-in-the-loop implementation in Nvidia Isaac Sim. The design employs three camera-equipped USVs and a YOLO-based detector to generate pixel bearings that close the loop with a multi-agent control policy. Throughout the experiments carried out, the proposed pipeline accomplishes sub-meter steady-state localization accuracy once a well-conditioned formation is established and remains operational with moderate perception dropouts.
Citation
Almozahmi, Mohammed Hamad Mothana, "Robust Incremental Bearing-Only Localizer," M.S. Thesis, Robotics, MBZUAI, 2026.
Source
Conference
Keywords
Maritime Search and Rescue, RIBOL, Unmanned Surface Vehicles (USVs), Computer Vision-Based Tracking, Sensor Fusion / Multi-Agent Estimation, Bearing-Only Localization