# CoReMatch：连续坐标迭代优化实现边缘端实时机器人特征匹配

> 原文链接：https://robotics.hellobike.com/papers/2026-11-04-2611-03932-cc5fefcd
> 频道：papers | 更新日期：2026-11-04 | 标签：auto, arxiv
> 核心摘要：提出CoReMatch框架，摒弃离散置信度图预测，在连续坐标空间迭代优化特征对应。实验显示其精度超越Efficient LoFTR，且在嵌入式边缘硬件上推理速度提升超10倍，为机器人SLAM与导航提供高效感知方案。

# CoReMatch：连续坐标迭代优化实现边缘端实时机器人特征匹配

::: v-pre
> 本文由 AI 根据公开内容整理，可能有误，请以原文为准：[CoReMatch: Recurrent Continuous Coordinate Refinement for Real-Time Robotic Feature Matching](https://arxiv.org/abs/2611.03932)（arXiv 2611.03932）

## 核心内容

针对现有半稠密匹配器受限于固定网格分辨率、难以兼顾精度与实时性的痛点，研究团队提出CoReMatch（Coordinate Refinement Match）循环框架。该方法直接在连续坐标空间进行对应关系细化，而非预测离散置信度图，通过跨迭代信息传播逐步提升估计精度，旨在解决机器人视觉定位、建图及自主导航中的核心感知瓶颈。

系统架构采用Deep Aggregated Transformer捕获全局上下文并迭代更新坐标，配合轻量级回归头进一步锐化定位精度，避免了高分辨率网格带来的算力开销。在室内外基准测试中，CoReMatch实现了精度与效率的强平衡：准确率优于Efficient LoFTR，同时在嵌入式边缘设备上运行速度提升10倍以上，验证了循环坐标细化在端侧实时特征匹配中的可扩展性。

## 关键看点

- **连续坐标优化**：摒弃固定网格离散预测，直接在连续空间迭代细化对应关系，突破预定义位置限制并降低高分辨率计算成本
- **网络架构设计**：集成Deep Aggregated Transformer捕获全局上下文，结合轻量级回归头锐化定位，通过循环机制跨迭代传播信息
- **边缘端性能突破**：在嵌入式硬件上推理速度较Efficient LoFTR提升超10倍，同时保持更高匹配精度，满足实时SLAM需求

## 工程与产业观察

该工作证明了在资源受限的边缘端，通过算法层面的连续坐标迭代替代暴力高分辨率网格搜索，是打破特征匹配“精度-速度”权衡的有效路径。对于具身智能终端部署而言，这种轻量化且高精度的感知前端能显著释放算力余量，为多模态融合与高层规划预留空间，具有重要的工程落地参考价值。

## 英文摘要 (Abstract)

Feature matching underpins many robotic perception tasks, including visual localization, mapping, SLAM, and autonomous navigation. Existing semi-dense matchers predict confidence on fixed grids, limiting correspondences to predefined locations and requiring costly high-resolution grids for precise localization. We present CoReMatch (Coordinate Refinement Match), a recurrent framework that refines correspondences directly in continuous coordinate space rather than predicting discrete confidence maps. CoReMatch progressively improves correspondence estimates by propagating information across iterations. At the coarse stage, a Deep Aggregated Transformer captures global context and iteratively updates correspondence coordinates, while a lightweight regression head further sharpens their localization. Experiments on indoor and outdoor benchmarks show that CoReMatch strikes a strong balance between accuracy and efficiency, outperforming Efficient LoFTR in accuracy while running more than 10x faster on embedded edge hardware. These results demonstrate that recurrent coordinate refinement offers an efficient and scalable approach to real-time feature matching for robotic perception.

作者：Robert J. Wang, Charles Ling, Boyu Wang
:::

## 原始参考来源 (Sources)

- [arXiv 2611.03932：CoReMatch: Recurrent Continuous Coordinate Refinement for Real-Time Robotic Feature Matching](https://arxiv.org/abs/2611.03932)