Loop Closure using AnyLoc Visual Place Recognition in DPV-SLAM

TL;DR

The paper tackles loop closure reliability in visual SLAM by replacing the traditional BoVW-based detector in DPV-SLAM with AnyLoc, a learning-based visual place recognition method. It introduces an adaptive similarity threshold and a complete loop-closure pipeline incorporating geometric verification and asynchronous pose graph optimization. The main contributions are integrating AnyLoc into DPV-SLAM, designing environment-aware thresholding, and demonstrating improved loop closure accuracy and robustness across indoor and outdoor scenes. The findings indicate practical improvements for real-world SLAM, with the caveat of substantial GPU resource requirements and potential future work on speedups and benchmarks.

Abstract

Loop closure is crucial for maintaining the accuracy and consistency of visual SLAM. We propose a method to improve loop closure performance in DPV-SLAM. Our approach integrates AnyLoc, a learning-based visual place recognition technique, as a replacement for the classical Bag of Visual Words (BoVW) loop detection method. In contrast to BoVW, which relies on handcrafted features, AnyLoc utilizes deep feature representations, enabling more robust image retrieval across diverse viewpoints and lighting conditions. Furthermore, we propose an adaptive mechanism that dynamically adjusts similarity threshold based on environmental conditions, removing the need for manual tuning. Experiments on both indoor and outdoor datasets demonstrate that our method significantly outperforms the original DPV-SLAM in terms of loop closure accuracy and robustness. The proposed method offers a practical and scalable solution for enhancing loop closure performance in modern SLAM systems.

Figures (14)

• Figure 1: DPV-SLAM pipeline with AnyLoc-based visual place recognition.
• Figure 2: Extraction of global descriptors using AnyLoc-VLAD-DINOv2.
• Figure 3: Corridor environment.
• Figure 4: 3D point cloud map and camera trajectory in the corridor environment (proposed method).
• Figure 5: Comparison of camera trajectories in the corridor environment.