Scholarlink

<p>To address the limitation of current automated road cleaning equipment, which struggles to collect nails, this paper presented a specialized robot system. This system’s function mainly consists of recognition, localization and retrieval of nails, of which recognition is the most important. To facilitate deployment of recognition algorithm on hardware while maintaining recognition accuracy and speed, a lightweight model, embedded with auxiliary detection head modules, coordinate convolution modules and upsampling operator modules was proposed. Then the model was deployed on NVIDIA Jetson Orin Nano along with the binocular stereo matching algorithm for simultaneous recognition and localization. Finally, a binocular localization-based electromagnetic retrieval system was designed and combined with the robot walking and control structure to construct the whole robot control system. In this system, images captured by binocular camera were transmitted to the NVIDIA JetSon Orin Nano device that deployed with the recognition and localization algorithms. Then, the obtained coordinates were fed back to the system to realize the retrieval of the nails. To ensure the learning ability of the proposed model, we collected and labeled 2100 color images for seven different working conditions and seven types of nails with the designed robot. The recognition experiments achieved an average accuracy of 96.9% in identifying nails, surpassing the performance of the backbone YOLOv5n by 2.1%. The localization error of nail coordinates within a 120°field of view across different scenarios was kept within 1.1 ~ 1.9&#xa0;cm. The error range ensures the localization precision and yields a 95.4% retrieval accuracy when combined with the electromagnetic retrieval system.</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

Robot system for accurate nail recognition, localization and retrieval

  • 期刊论文

  • 发表日期：

    2025年9月26日

查看全文
  • Haijian Wang,
  • Ziliang Hu,
  • Xuemei Zhao

摘要

To address the limitation of current automated road cleaning equipment, which struggles to collect nails, this paper presented a specialized robot system. This system’s function mainly consists of recognition, localization and retrieval of nails, of which recognition is the most important. To facilitate deployment of recognition algorithm on hardware while maintaining recognition accuracy and speed, a lightweight model, embedded with auxiliary detection head modules, coordinate convolution modules and upsampling operator modules was proposed. Then the model was deployed on NVIDIA Jetson Orin Nano along with the binocular stereo matching algorithm for simultaneous recognition and localization. Finally, a binocular localization-based electromagnetic retrieval system was designed and combined with the robot walking and control structure to construct the whole robot control system. In this system, images captured by binocular camera were transmitted to the NVIDIA JetSon Orin Nano device that deployed with the recognition and localization algorithms. Then, the obtained coordinates were fed back to the system to realize the retrieval of the nails. To ensure the learning ability of the proposed model, we collected and labeled 2100 color images for seven different working conditions and seven types of nails with the designed robot. The recognition experiments achieved an average accuracy of 96.9% in identifying nails, surpassing the performance of the backbone YOLOv5n by 2.1%. The localization error of nail coordinates within a 120°field of view across different scenarios was kept within 1.1 ~ 1.9 cm. The error range ensures the localization precision and yields a 95.4% retrieval accuracy when combined with the electromagnetic retrieval system.