EUVIMEDEUVIMED entdecken
Open Access · CC BY-NC-ND 4.0

Lightweight human-following for agricultural robots via low-viewpoint lower-body detection and latency-compensated motion prediction

Hyun-Sik Son, Deok-Keun Kim, Seung-Hwan Yang

ICT Express · 2026-11

DOI: 10.1016/j.icte.2026.10.016 ↗

Kurz gefasst

Fachartikel in der Zeitschrift ICT Express, 2026-11. Verfasst von Hyun-Sik Son und 2 weiteren Autor:innen. Die Arbeit nennt 22 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY-NC-ND 4.0).

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

This paper presents a lightweight human-following system for agricultural harvest-assistance robots, combining: (1) a lower-body-specific You Only Look Once v5-small (YOLOv5s) detection pipeline trained on 3286 images (99.4% mAP@0.5), adapted to a 0.2 m camera-mounting constraint and deployed on a Jetson Nano; and (2) a latency-compensated motion prediction (LCMP) algorithm. Indoor experiments demonstrate continuous following at 0.8–1.2 m/s. At 1.0 m/s, LCMP reduces distance root mean square error (RMSE) by 46% and variance by 72%; turning tracking loss decreases from 79%–83% to 10%–16% above 0.8 m/s. A greenhouse trial on a commercial platform provided preliminary evidence of feasibility.

Zitieren

APA

Son, H., Kim, D., & Yang, S. (2026). Lightweight human-following for agricultural robots via low-viewpoint lower-body detection and latency-compensated motion prediction. ICT Express. https://doi.org/10.1016/j.icte.2026.10.016

Vancouver

Son H, Kim D, Yang S. Lightweight human-following for agricultural robots via low-viewpoint lower-body detection and latency-compensated motion prediction. ICT Express. 2026. doi:10.1016/j.icte.2026.10.016

BibTeX
@article{Son2026,
  author = {Son, Hyun-Sik and Kim, Deok-Keun and Yang, Seung-Hwan},
  title = {Lightweight human-following for agricultural robots via low-viewpoint lower-body detection and latency-compensated motion prediction},
  journal = {ICT Express},
  year = {2026},
  doi = {10.1016/j.icte.2026.10.016}
}
Kostenlos starten →