Scholarlink

<p>This article highlights new developments in computer vision and robotics. Thanks to learning-based methods, new solutions have emerged for challenges such as recognition, pose estimation, and grasping of novel, previously unseen objects. By leveraging learned geometric features, it is possible to recognize new object instances and assign them to classes that are similar to but different from known ones, thereby enabling autonomous robotic grasping. Object detection and pose estimation are also feasible for transparent objects, and their fill levels can be reliably determined. To ensure that detected objects are indeed present and their positions accurate, the method of object position verification is applied. This also helps correct potential hallucinations produced by learned methods. Finally, we present approaches that allow not only the transfer of points but also entire processing paths from known objects to new ones.</p>

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

Methoden zur raschen Erkennung neuer Objekte in Farb- und Tiefenbildern

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2025年10月15日

查看全文
  • Markus Vincze,
  • Jean-Baptiste Weibel,
  • Philipp Ausserlechner,
  • Peter Hönig,
  • Paolo Sebeto,
  • Anish Pratheepkumar,
  • Hrishikesh Gupta,
  • Stefan Thalhammer

摘要

This article highlights new developments in computer vision and robotics. Thanks to learning-based methods, new solutions have emerged for challenges such as recognition, pose estimation, and grasping of novel, previously unseen objects. By leveraging learned geometric features, it is possible to recognize new object instances and assign them to classes that are similar to but different from known ones, thereby enabling autonomous robotic grasping. Object detection and pose estimation are also feasible for transparent objects, and their fill levels can be reliably determined. To ensure that detected objects are indeed present and their positions accurate, the method of object position verification is applied. This also helps correct potential hallucinations produced by learned methods. Finally, we present approaches that allow not only the transfer of points but also entire processing paths from known objects to new ones.