HeLiOS: Heterogeneous LiDAR Place Recognition via Overlap-based Learning and Local Spherical Transformer

TL;DR

The paper tackles heterogeneous LiDAR place recognition by introducing HeLiOS, a single-branch network that learns robust global descriptors through locality-aware local feature extraction, overlap-based clustering, and an OT-based aggregation. Training is guided by overlap-based data mining and a guided-triplet loss with adaptive margins, addressing weaknesses of distance-based mining and fixed-margin triplet approaches. Empirical results on public datasets (NCLT, MulRan, HeLiPR) show state-of-the-art performance for cross-sensor and long-term recognition, with lightweight variants preserving strong accuracy. The work provides an open-source implementation and establishes a foundation for cross-LiDAR place recognition in real-world robotics applications.

Abstract

LiDAR place recognition is a crucial module in localization that matches the current location with previously observed environments. Most existing approaches in LiDAR place recognition dominantly focus on the spinning type LiDAR to exploit its large FOV for matching. However, with the recent emergence of various LiDAR types, the importance of matching data across different LiDAR types has grown significantly-a challenge that has been largely overlooked for many years. To address these challenges, we introduce HeLiOS, a deep network tailored for heterogeneous LiDAR place recognition, which utilizes small local windows with spherical transformers and optimal transport-based cluster assignment for robust global descriptors. Our overlap-based data mining and guided-triplet loss overcome the limitations of traditional distance-based mining and discrete class constraints. HeLiOS is validated on public datasets, demonstrating performance in heterogeneous LiDAR place recognition while including an evaluation for long-term recognition, showcasing its ability to handle unseen LiDAR types. We release the HeLiOS code as an open source for the robotics community at https://github.com/minwoo0611/HeLiOS.

Figures (7)

• Figure 1: HeLiOS utilizes overlap for mining and the loss function. Conventional distance-based mining might lead to incorrect pairings, such as blue-yellow circles (close distance, no overlap) or black-yellow circles (far apart, overlap). Incorrect pairings can negatively impact both the training process and overall performance.
• Figure 2: The overall pipeline of HeLiOS. HeLiOS voxelizes point clouds $\mathbf{P}$ from heterogeneous LiDARs, which are then processed through a shared feature extraction network with sparse convolution and a spherical transformer. Local features $\mathbf{F}$ are aggregated into global descriptors using GeM and SALAD to produce similar descriptors, which are subsequently utilized for training and evaluation.
• Figure 3: Local spherical window for applying multi-head attention in heterogeneous LPR. Due to the differing global distribution when scanning the entire scene with different LiDARs, training attention is challenging. Distribution is closer within smaller, localized windows, enabling effective attention learning.
• Figure 4: (a) Overlap calculation to illustrate \ref{['eq:overlap']} as a diagram. (b) HeLiOS computes overlap for different LiDARs in 3D space. In contrast, OverlapNet misrepresents real-world overlap even if the LiDARs are in the same location, as their overlap occurs only when each point falls into the same pixel in a range image.
• Figure 5: PR curves with heterogeneous LiDARs. The title of each curve represents the database from Roundabout01 and the query from Roundabout02. HeLiOS surpasses other methods regardless of the size of the descriptor.