On Motion Blur and Deblurring in Visual Place Recognition

TL;DR

This work addresses the underexplored problem of motion blur in Visual Place Recognition (VPR) by introducing Blurry Places, a benchmark built from high-frame-rate video to synthesize motion blur across nine intensities $L$ on three outdoor routes. It systematically evaluates a range of VPR methods with and without three deblurring techniques (DeblurGANv2, FFTFormer, GShift-Net) and a blur-detection–driven adaptive deblurring approach. Key findings show that motion blur degrades VPR performance across methods, deblurring can yield substantial gains especially at high blur levels, and adaptive deblurring offers a trade-off between accuracy and compute in realistic scenarios. The results motivate future development of blur-aware, computationally efficient VPR and deblurring pipelines suitable for robotic platforms and low-power devices.

Abstract

Visual Place Recognition (VPR) in mobile robotics enables robots to localize themselves by recognizing previously visited locations using visual data. While the reliability of VPR methods has been extensively studied under conditions such as changes in illumination, season, weather and viewpoint, the impact of motion blur is relatively unexplored despite its relevance not only in rapid motion scenarios but also in low-light conditions where longer exposure times are necessary. Similarly, the role of image deblurring in enhancing VPR performance under motion blur has received limited attention so far. This paper bridges these gaps by introducing a new benchmark designed to evaluate VPR performance under the influence of motion blur and image deblurring. The benchmark includes three datasets that encompass a wide range of motion blur intensities, providing a comprehensive platform for analysis. Experimental results with several well-established VPR and image deblurring methods provide new insights into the effects of motion blur and the potential improvements achieved through deblurring. Building on these findings, the paper proposes adaptive deblurring strategies for VPR, designed to effectively manage motion blur in dynamic, real-world scenarios.

Figures (5)

• Figure 1: Our main contributions include a new benchmark with motion blur and scene variation, alongside an analysis of VPR performance under motion blur, deblurring, and adaptive deblurring. Examples show incorrect (red) and correct (green) matches before and after deblurring.
• Figure 2: A place captured in Luzzara in three traverses and blur intensities.
• Figure 3: AUC variation for increasing motion blur intensity. The same traverse is used as reference and query to exclude all the others appearance changes.
• Figure 4: AUC variation for increasing motion blur intensity combined with other appearance changes.
• Figure 5: Heatmaps showing the difference in AUC performance after deblurring the other datasets (rows) with each deblurring method (columns), with all VPR method and blur intensity combinations.