강인한 위치추정을 위한 센서융합 및 사전정보 활용 연구 = Robust Localization Approach with Sensor Fusion and Prior Information

다국어 초록 (Multilingual Abstract)

Title: Robust Localization Approach with Sensor Fusion and Prior Information Reliable localization over wide areas is required for the autonomous opera- tion of mobile platforms in real-world environments. However, GNSS accuracy and stability can be degraded due to urban canyons, occlusion, and multipath effects. Conversely, visual SLAM methods face limitations in large-scale applicability due to the burdens of prior exploration and map construction. To achieve robust local- ization for mobile platforms, this thesis proposes and experimentally validates two complementary approaches: sensor fusion and prior information. From the sensor fusion perspective, we address the issue of GNSS vertical accuracy degradation by proposing the barometric velocity correction (BVC) method, which interprets baro- metric measurements as vertical velocity information rather than direct altitude values. BVC was designed to be effectively fused with GNSS within a Kalman filter framework while mitigating the effects of barometric drift. Through synthetic and real-world experiments, we confirmed that the proposed method significantly im- proved altitude stability and 3D positioning accuracy compared to GNSS-only and conventional barometer fusion methods. From the prior information perspective, we propose TileLoc, a VPR-based UAV global localization framework that references public web tile maps. While web tile maps offer wide coverage and high accessi- bility, direct matching is challenging due to viewpoint differences, limited overlap, and rotation and scale mismatches with UAV imagery. To address this, we designed a lightweight global localization method combining a multi-zoom overlapped tile database, rotated query batches, and a sequence-based voting mechanism. Exper- imental results using the UAV-VisLoc dataset confirmed that each module incre- mentally improved performance, enabling robust and stable global localization in diverse environments.

목차 (Table of Contents)

• Summary i
• List of Tables v
• List of Figures x
• Chapter 1. Introduction 1
• Chapter 2. Sensor Fusion Approach 3
• 2.1 Motivation 3
• 2.2 Related Works 4
• 2.3 BVC (Barometric Velocity Correction) 8
• 2.4 Experiments with Synthetic Data 18
• 2.5 Experiments with Real Data 27
• 2.6 Summary and Future Works 33
• Chapter 3. Prior Information Approach 34
• 3.1 Motivation 34
• 3.2 Related Works 35
• 3.3 Preliminary Analysis 37
• 3.4 Visual Place Recognition with Sequential Filtering 42
• iii
• 3.5 Experiments 46
• 3.6 Summary and Future Works 48
• Chapter 4. Conclusion 52
• 4.1 Summary 52
• 4.2 Perspectives 52
• References 54
• 국문초록 63
• 감사의 글 64
• iv