Voronoi-Based Second-Order Descriptor with Whitened Metric in LiDAR Place Recognition
Jaein Kim, HEE BIN YOO, Dong-Sig Han, Byoung-Tak Zhang
AI summary
Problem
Existing LiDAR place recognition models rely on first-order pooling methods that fail to capture higher-order feature interactions and use inappropriate normalization techniques that distort the natural cluster structure of global descriptors.
Approach
The authors propose a novel pooling method that aggregates local descriptors into a second-order matrix and applies instance-level ZCA whitening within each Voronoi cell to enable accurate Mahalanobis distance measurement while maintaining numerical stability during end-to-end training.
Key results
- State-of-the-art recall on Oxford RobotCar and Wild-Places benchmarks
- Outperforms first-order baselines even with reduced descriptor dimensions
- Superior descriptor space metrization evidenced by MAP@K and ROC curves
- Stable end-to-end training via Rao-Blackwell Ledoit-Wolf covariance shrinkage
Why it matters
Enhances the robustness and accuracy of LiDAR-based loop closure and global localization for autonomous navigation in diverse urban and natural environments.
Abstract
The pooling layer plays a vital role in aggre- gating local descriptors into the metrizable global descriptor in the LiDAR Place Recognition (LPR). In particular, the second-order pooling is capable of capturing higher-order interactions among local descriptors. However, its existing methods in the LPR adhere to conventional implementations and post-normalization, and incur the descriptor unsuitable for Euclidean distancing. Based on the recent interpretation that associates NetVLAD with the second-order statistics, we propose to integrate second-order pooling with the inductive bias from Voronoi cells. Our novel pooling method aggregates local descriptors to form the second-order matrix and whitens the global descriptor to implicitly measure the Mahalanobis distance while conserving the cluster property from Voronoi cells, addressing its numerical instability during learning with diverse techniques. We demonstrate its performance gains through the experiments conducted on the Oxford Robotcar and Wild-Places benchmarks and analyze the numerical effect of the proposed whitening algorithm.