Evaluation of range sensing-based place recognition for long-term urban localization

Abstract

Place recognition is a critical capability for autonomous vehicles. It matches current sensor data with a pre-built database to provide coarse localization results. However, the effectiveness of long-term place recognition may be degraded by environment changes, such as seasonal or weather changes. To have a deep understanding of this issue, we conduct a comprehensive evaluation study on several state-of-the-art range sensing-based (i.e., LiDAR and radar) place recognition methods on the Borease dataset, which encapsulates long-term localization scenarios with stark seasonal variations and adverse weather conditions. In addition, we design a novel metric to evaluate the influence of matching thresholds on place recognition performance for long-term localization. Our results and findings provide fresh insights to the community and potential directions for future study.