Scholarlink

<p>Robotic systems for automated grasping on conveyor Lines require precise 3D vision for object detection, localization, and manipulation in dynamic environments. Traditional robotic arms lack adaptability due to fixed programming and static sensors. Advances in deep learning, Like the YOLO model, and improved depth sensing have enhanced robotic perception. However, achieving accurate 3D localization with low-cost depth cameras remains challenging due to noise and depth estimation errors affecting grasping precision. To address this challenge, we propose an integrated 3D vision system combining the YOLOv8 object detection model with an Intel RealSense D435 depth camera to achieve real-time and accurate 3D localization. A 4-degree-of-freedom (DOF) SCARA robot, controlled by a programmable logic controller (PLC), is used to grasp objects with high precision. The YOLOv8 model, trained on a custom dataset tailored for production Line settings, achieves an object detection accuracy of 97.3% with an average processing speed of 31.1&#xa0;ms per frame. Depth-filtering techniques are applied to enhance depth image quality, significantly reducing noise and improving localization accuracy. Experimental results show that applying all depth filters together reduces the subpixel root mean square (RMS) error by over 78% at a 1&#xa0;m distance and nearly 82% at 1.7&#xa0;m, demonstrating substantial improvements in depth estimation. The optimized depth data contribute to an overall localization accuracy of 3.2&#xa0;mm, leading to more precise object grasping. Grasping experiments validate the system’s performance, showing reliable handling of objects with varying shapes and positions.</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

Development of a 3D Vision Robot System to Grasp Objects on the Conveyor Using Artificial Intelligence

  • 期刊论文

  • 发表日期：

    2025年9月16日

查看全文
  • Vo Duy Cong,
  • Le Hoai Phuong

摘要

Robotic systems for automated grasping on conveyor Lines require precise 3D vision for object detection, localization, and manipulation in dynamic environments. Traditional robotic arms lack adaptability due to fixed programming and static sensors. Advances in deep learning, Like the YOLO model, and improved depth sensing have enhanced robotic perception. However, achieving accurate 3D localization with low-cost depth cameras remains challenging due to noise and depth estimation errors affecting grasping precision. To address this challenge, we propose an integrated 3D vision system combining the YOLOv8 object detection model with an Intel RealSense D435 depth camera to achieve real-time and accurate 3D localization. A 4-degree-of-freedom (DOF) SCARA robot, controlled by a programmable logic controller (PLC), is used to grasp objects with high precision. The YOLOv8 model, trained on a custom dataset tailored for production Line settings, achieves an object detection accuracy of 97.3% with an average processing speed of 31.1 ms per frame. Depth-filtering techniques are applied to enhance depth image quality, significantly reducing noise and improving localization accuracy. Experimental results show that applying all depth filters together reduces the subpixel root mean square (RMS) error by over 78% at a 1 m distance and nearly 82% at 1.7 m, demonstrating substantial improvements in depth estimation. The optimized depth data contribute to an overall localization accuracy of 3.2 mm, leading to more precise object grasping. Grasping experiments validate the system’s performance, showing reliable handling of objects with varying shapes and positions.