New benchmark reveals motion blur impact on robot visual place recognition

Visual Place Recognition (VPR) is critical for mobile robot localization, but most benchmarks ignore motion blur—a problem that arises not only in fast motion but also in low-light conditions requiring longer exposures. The paper by Ismagilov et al. fills this gap by introducing a comprehensive benchmark with three datasets that cover a wide range of motion blur intensities. The authors evaluate several well-established VPR and image deblurring methods, providing new insights into how blur degrades recognition accuracy and how deblurring can recover performance.

Building on these experiments, the paper proposes adaptive deblurring strategies for VPR that dynamically select the best deblurring approach based on the observed blur level. This enables robots to maintain reliable localization in dynamic, real-world environments—such as autonomous vehicles driving through tunnels or drones flying at dusk. The work was accepted to IEEE Robotics & Automation Letters and offers a practical foundation for robust, blur-agnostic visual navigation systems.