Role of Deep Learning in Loop Closure Detection for Visual and Lidar SLAM: A Survey.

Area of Science:

• Robotics
• Computer Vision
• Artificial Intelligence

Background:

• Loop closure detection is essential for Simultaneous Localization and Mapping (SLAM) to ensure accurate robot positioning and consistent global map generation.
• Existing methods vary in view acquisition, scene representation, and matching strategies, presenting limitations in real-world applications.

Purpose of the Study:

• To provide a comprehensive survey of loop closure detection algorithms for visual and Lidar SLAM.
• To present a taxonomy and comparison of state-of-the-art deep learning-based loop detection algorithms.
• To identify challenges in conventional methods and review deep learning solutions for long-term autonomy.

Main Methods:

• Literature review of existing loop closure detection algorithms for visual and Lidar SLAM.
• Taxonomy and detailed comparison of deep learning-based loop detection algorithms.
• Analysis of challenges in conventional approaches and review of deep learning methods addressing them.

Main Results:

• A thorough study of loop closure detection literature, highlighting insights and limitations.
• A taxonomy and comparative analysis of deep learning-based loop detection algorithms.
• Identification of key challenges in traditional methods and how deep learning tackles them for robust performance.

Conclusions:

• Deep learning-based methods offer promising solutions to overcome limitations of conventional loop closure detection in SLAM.
• Addressing challenges like changing weather, lighting, viewpoint, and occlusion is key for long-term robotic autonomy.
• Further research into open challenges and future directions is needed for advancing robust SLAM systems.