[Paper Review] Vision-based localization methods under GPS-denied conditions

Zihao Lu, Fei Liu+1 more|arXiv (Cornell University)|Nov 22, 2022

Robotics and Sensor-Based Localization4 citations

TL;DR

This paper reviews vision-based localization methods in GPS-denied environments, categorizing them into Relative Vision Localization (RVL) and Absolute Vision Localization (AVL). It analyzes optical flow in feature-based Visual Odometry, optimization and EKF-based VSLAM, and offline map registration with lane detection, offering performance comparisons and future research directions for robust visual localization without GPS.

ABSTRACT

This paper reviews vision-based localization methods in GPS-denied environments and classifies the mainstream methods into Relative Vision Localization (RVL) and Absolute Vision Localization (AVL). For RVL, we discuss the broad application of optical flow in feature extraction-based Visual Odometry (VO) solutions and introduce advanced optical flow estimation methods. For AVL, we review recent advances in Visual Simultaneous Localization and Mapping (VSLAM) techniques, from optimization-based methods to Extended Kalman Filter (EKF) based methods. We also introduce the application of offline map registration and lane vision detection schemes to achieve Absolute Visual Localization. This paper compares the performance and applications of mainstream methods for visual localization and provides suggestions for future studies.

Motivation & Objective

• To systematically review and classify vision-based localization methods applicable in GPS-denied environments.
• To analyze the strengths and limitations of Relative Vision Localization (RVL) techniques, particularly those based on optical flow and Visual Odometry.
• To examine Absolute Vision Localization (AVL) approaches, including EKF-based and optimization-driven VSLAM, and offline map registration methods.
• To evaluate the integration of lane vision detection for improving absolute localization accuracy in structured environments.
• To provide comparative insights into performance and applicability of mainstream methods and suggest future research directions.

Proposed method

• Classifies vision-based localization into two main categories: Relative Vision Localization (RVL) and Absolute Vision Localization (AVL).
• Reviews optical flow estimation techniques used in feature extraction for Visual Odometry (VO) in RVL.
• Analyzes optimization-based and Extended Kalman Filter (EKF)-based approaches in Visual Simultaneous Localization and Mapping (VSLAM) for AVL.
• Examines offline map registration techniques that align visual observations with pre-built maps for absolute localization.
• Introduces lane vision detection schemes as a supplementary method to enhance localization accuracy in road environments.
• Synthesizes performance comparisons across methods, focusing on accuracy, robustness, and computational efficiency.

Experimental results

Research questions

• RQ1 How do RVL and AVL methods differ in their approach to visual localization under GPS-denied conditions?
• RQ2 What role does optical flow play in improving feature-based Visual Odometry for relative localization?
• RQ3 How do EKF-based and optimization-based VSLAM techniques compare in terms of accuracy and stability for absolute localization?
• RQ4 In what ways can offline map registration and lane detection enhance the performance of absolute visual localization?
• RQ5 What are the key performance trade-offs among current vision-based localization methods in GPS-denied scenarios?

Key findings

• Optical flow-based methods in RVL demonstrate strong performance in dynamic and texture-rich environments due to dense motion estimation.
• EKF-based VSLAM offers real-time performance with moderate accuracy, while optimization-based methods achieve higher localization precision at the cost of increased computational load.
• Offline map registration significantly improves localization accuracy by enabling global consistency through pre-mapped visual references.
• Lane vision detection schemes enhance AVL performance in structured road environments by providing geometric constraints and semantic context.
• The integration of multiple visual cues—such as optical flow, map registration, and lane detection—leads to more robust and accurate localization than single-method approaches.
• The paper identifies a research gap in long-term localization stability and generalization across diverse environments, suggesting future work should focus on hybrid learning-based and geometric methods.