XPRESS: X-Band Radar Place Recognition via Elliptical Scan Shaping

TL;DR

The paper tackles the problem of robust place recognition for autonomous maritime navigation using X-band radar, which suffers from low resolution and rotational distortion. It introduces XPRESS, a two-stage framework that first converts radar clusters into an Elliptical Scan representation to suppress fluctuations, then performs fast, rotationally invariant matching in a polar histogram space via KD-tree retrieval. Key contributions include the first X-band radar–specific PR method for maritime settings, a cluster-count–based candidate pruning strategy, and a fully metric descriptor with ablation-informed parameter tuning, demonstrated across MOANA, Pohang Canal, and a private dataset with intra- and inter-session PR. The results show improved robustness and faster retrieval relative to state-of-the-art W-band PR methods, highlighting XPRESS’s potential to enable radar-only SLAM and long-term maritime autonomy in GNSS-denied environments.

Abstract

X-band radar serves as the primary sensor on maritime vessels, however, its application in autonomous navigation has been limited due to low sensor resolution and insufficient information content. To enable X-band radar-only autonomous navigation in maritime environments, this paper proposes a place recognition algorithm specifically tailored for X-band radar, incorporating an object density-based rule for efficient candidate selection and intentional degradation of radar detections to achieve robust retrieval performance. The proposed algorithm was evaluated on both public maritime radar datasets and our own collected dataset, and its performance was compared against state-of-the-art radar place recognition methods. An ablation study was conducted to assess the algorithm's performance sensitivity with respect to key parameters.

Figures (9)

• Figure 1: XPRESS proposes a fast and robust PR method for maritime environments using X-band radar. By leveraging radar cluster information and applying intentional elliptical degradation, the proposed approach minimizes the influence of dynamic elements and fluctuations of static objects.
• Figure 2: The concept of the proposed algorithm lies in the intentional degradation of radar data. Although static objects are detected, their appearance fluctuates across frames. To address this, radar data is first clustered using connected component labeling. Each labeled component is then approximated by an ellipse. The elliptical data is transformed into polar coordinates, from which a polar histogram is generated to enable KD-Tree-based retrieval. During descriptor matching, candidates are initially filtered based on the number of clusters, followed by estimation of revisitness among the selected candidates.
• Figure 3: In coastal areas, while the density of anchored vessels remains relatively stable, their individual positions vary consistently. Considering the vessel density within specific regions, rather than relying on exact vessel positions, is advantageous for handling dynamic elements in X-band radar images.
• Figure 4: X-band radar data exhibits fluctuations, the same objects across consecutive frames appear with varying cluster boundaries. To mitigate the impact of this variability, the radar data is degraded into elliptical representations (top-left). The number of resulting clusters (top-right) serves as a criterion for initial candidate selection, and the histogram derived from the polar-transformed data is used as a descriptor for location retrieval (bottom).
• Figure 5: The left image illustrates our sensor set and the rights are the routes for the evalutation. To evaluate the proposed algorithm, we utilized two public datasets and additionally acquired our own dataset in the same geographic region as the MOANA dataset. Our dataset is divided into two sequences, Coastal and Complex, which are used to assess intra-session and long-term inter-session PR, respectively.