# TRACKGRAPH：通过图像空间跟踪实现在线开放词汇3D场景图

> 原文链接：https://robotics.hellobike.com/papers/2026-09-25-2609-31005
> 频道：papers | 更新日期：2026-09-25 | 标签：arxiv, cs.RO
> 核心摘要：TRACKGRAPH通过图像空间跟踪减少逐帧分割与VL推理，在多个数据集、四足机器人和无人机数据上验证在线开放词汇3D场景图。

# TRACKGRAPH：通过图像空间跟踪实现在线开放词汇3D场景图

::: v-pre
> 原文链接：[TRACKGRAPH: Online Open-Vocabulary 3D Scene Graphs via Image-Space Tracking](https://arxiv.org/abs/2609.31005)（arXiv 2609.31005 · 发布日期：2026-09-25）

**TRACKGRAPH: Online Open-Vocabulary 3D Scene Graphs via Image-Space Tracking**

TRACKGRAPH通过图像空间跟踪减少逐帧分割与VL推理，在多个数据集、四足机器人和无人机数据上验证在线开放词汇3D场景图。

## 核心要点

- 现有开放词汇3D地图常需逐帧分割、关联持久3D段，并频繁进行昂贵VL推理。
- 方法在图像流中维护短期2D掩码身份，再将跟踪掩码融合进层级3D场景图。
- 稀疏关键帧计算FastSAM掩码和CLIP特征，中间用DINOv3高频传播掩码。
- 在Replica、ScanNet++、HM3D上结果具竞争力；同A100上更快且显存更少。

**验证范围与局限**：验证范围包括Replica、ScanNet++、HM3D、真实四足机器人部署和记录的无人机数据。

## 英文摘要 (Abstract)

Open-vocabulary 3D maps enable robots to reason about previously unknown environments using natural language. However, existing systems typically segment every incoming image, associate detections with persistent 3D segments, and frequently perform costly Vision-Language (VL) inference. We present TRACKGRAPH, an online open-vocabulary system that maintains short-term 2D mask identity directly in the image stream before fusing segments into 3D. FastSAM masks and CLIP features are computed at sparse keyframes, while dense DINOv3 features are used to propagate masks at a high rate in between. The resulting tracked masks are fused into a class-agnostic 3D segment layer within a hierarchical scene graph, with 3D association handling tracking interruptions and long-term revisits. Compact multi-view CLIP embeddings enable open-vocabulary retrieval. Across Replica, ScanNet++, and HM3D, TRACKGRAPH achieves competitive open-vocabulary segmentation and retrieval against state-of-the-art mapping methods, including the highest synonym frequency on Replica (0.50). On the same NVIDIA A100, it is 1.7x faster and uses 3.3x less GPU memory than ViT-H OVI-MAP. Real-world quadruped deployments demonstrate onboard scene graph construction and object search at 7.5Hz, while recorded drone data is used to test the method under aerial viewpoints.

作者：Peder Borge Hellesylt, Albert Gassol Puigjaner, Kostas Alexis, Annette Stahl
:::

## 原始参考来源 (Sources)

- [arXiv 2609.31005：TRACKGRAPH: Online Open-Vocabulary 3D Scene Graphs via Image-Space Tracking](https://arxiv.org/abs/2609.31005)