Scholarlink

<p>This study presents a robust and efficient markerless deep learning–based framework for real-time position and heading measurement tracking of cyborg insects operating in unstructured and obstacle-present environments. By integrating a lightweight YOLO detector augmented with three pose estimation strategies: YOLO-based keypoints, a CNN heatmap model, and DeepLabCut, the system could reliably track small cyborg insect targets occupying approximately ~ 0.13% of the image frame using a single overhead web camera. Experiment results with single and two cyborg cockroaches demonstrated accurate trajectory measurement, stable identity preservation, and reliable heading estimation in both obstacle-free and obstacle-present arenas. Quantitative results showed that the YOLO-based pose estimator achieved a favorable balance of accuracy, robustness, and low inference latency (32&#xa0;ms), demonstrating improved performance compared to CNN heatmap-based and DLC approaches for continuous and simultaneous tracking and measurement. The proposed markerless tracking approach removed the need for reflective marker frames, thereby avoiding the additional payload imposed on the cyborg insect. No obstacle entanglement with obstacle/wall was observed in the conducted experiments when using the proposed markerless tracking approach, in contrast to prior marker-based setups. The real-time integration of markerless tracking and wireless backpack enabled synchronized locomotion monitoring at 23 FPS, showing a clear correlation between commands and insect kinematics. The proposed system supported natural locomotion of backpack-equipped cyborg insects and demonstrated potential suitability for real-time locomotion tracking experiments in controlled laboratory arenas.</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

Markerless deep-learning–based position and heading measurement for cyborg insects in unstructured environments

  • 期刊论文

  • 发表日期：

    2026年7月28日

查看全文
  • Habib Ja’far Nuur,
  • Mochammad Ariyanto,
  • Hafiz Akbar Simanjorang,
  • Ade Kurniawan,
  • M. Munadi,
  • Rifky Ismail,
  • Keisuke Morishima

摘要

This study presents a robust and efficient markerless deep learning–based framework for real-time position and heading measurement tracking of cyborg insects operating in unstructured and obstacle-present environments. By integrating a lightweight YOLO detector augmented with three pose estimation strategies: YOLO-based keypoints, a CNN heatmap model, and DeepLabCut, the system could reliably track small cyborg insect targets occupying approximately ~ 0.13% of the image frame using a single overhead web camera. Experiment results with single and two cyborg cockroaches demonstrated accurate trajectory measurement, stable identity preservation, and reliable heading estimation in both obstacle-free and obstacle-present arenas. Quantitative results showed that the YOLO-based pose estimator achieved a favorable balance of accuracy, robustness, and low inference latency (32 ms), demonstrating improved performance compared to CNN heatmap-based and DLC approaches for continuous and simultaneous tracking and measurement. The proposed markerless tracking approach removed the need for reflective marker frames, thereby avoiding the additional payload imposed on the cyborg insect. No obstacle entanglement with obstacle/wall was observed in the conducted experiments when using the proposed markerless tracking approach, in contrast to prior marker-based setups. The real-time integration of markerless tracking and wireless backpack enabled synchronized locomotion monitoring at 23 FPS, showing a clear correlation between commands and insect kinematics. The proposed system supported natural locomotion of backpack-equipped cyborg insects and demonstrated potential suitability for real-time locomotion tracking experiments in controlled laboratory arenas.