Scholarlink

This paper introduces a Robot Operating System 2 (ROS2) software pipeline, designed for the Jetson Orin Nano, that performs open-vocabulary object detection and subsequent 3D reconstruction from RGB images. The pipeline operates in two stages: an object detection module filters images for target objects, and a Multi-View Stereo (MVS) module then reconstructs the object in 3D. Notably, this MVS module employs an unconstrained multi-view stereo vision transformer, eliminating the need for explicit camera parameters. Due to limited memory, the pipeline used low-resolution images, compromising visual reconstruction quality and camera pose estimation accuracy. Nevertheless, the pipeline successfully identified semantically relevant objects and generated valuable 3D reconstructions for robotics applications.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

ROS2 Pipeline for Open-Vocabulary Object Detection and 3D Reconstruction on Jetson Orin Nano

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Sergio Oliveros,
  • Rubén Manrique

摘要

This paper introduces a Robot Operating System 2 (ROS2) software pipeline, designed for the Jetson Orin Nano, that performs open-vocabulary object detection and subsequent 3D reconstruction from RGB images. The pipeline operates in two stages: an object detection module filters images for target objects, and a Multi-View Stereo (MVS) module then reconstructs the object in 3D. Notably, this MVS module employs an unconstrained multi-view stereo vision transformer, eliminating the need for explicit camera parameters. Due to limited memory, the pipeline used low-resolution images, compromising visual reconstruction quality and camera pose estimation accuracy. Nevertheless, the pipeline successfully identified semantically relevant objects and generated valuable 3D reconstructions for robotics applications.