Design and Implementation of an AI-Integrated 3D Vision Conveyor System for Automated Inspection
Advanced Devices & Instrumentation · 01.01.2026 · 7
DOI: 10.34133/adi.0201 ↗Kurz gefasst
Fachartikel in der Zeitschrift Advanced Devices & Instrumentation, online erschienen am 17.07.2026. Verfasst von Cong Duy Vo und 2 weiteren Autor:innen. Die Arbeit nennt 39 Literaturangaben.
Abstract (Originalsprache)
The rapid expansion of e-commerce and Industry 4.0 logistics demands automated inspection systems that ensure accuracy, efficiency, and full data traceability. Conventional approaches based on dynamic weighing or 2-dimensional (2D) scanning often suffer from high error rates and the inability to provide comprehensive 3D information. This paper presents the design and implementation of an artificial intelligence (AI)-integrated 3D vision conveyor system for automated carton inspection, combining static weighing, multiview depth sensing, and radio frequency identification-based identification. A load cell module ensures high-precision weight measurement, while 3 orthogonally arranged RGB-D cameras capture carton dimensions and volume. To evaluate detection robustness, 2 state-of-the-art deep learning models, You Only Look Once version 12 (YOLOv12) and RT-DETR (Real time DEtection TRansformer), are trained on the same dataset and compared in terms of accuracy, computational cost, and real-time performance. Depth images are refined using a combination of spatial filters and random sample consensus-based quadratic surface fitting, reducing sensor noise and achieving millimeter-level accuracy in size estimation. An radio frequency identification module provides unique carton identification and seamless data synchronization with warehouse management systems, Enterprise Resource Planning, and Manufacturing Execution System. The software architecture, developed in Python with Qt Designer, enables real-time control, visualization, and historical data management, while supporting flexible integration of additional modules such as dashboards or predictive analytics. Experimental validation demonstrates that both detection models achieve mAP@0.5 of ~0.99 and mAP@0.5–0.95 of ~0.93, with YOLOv12 providing superior inference speed for real-time applications, while RT-DETR yields slightly higher accuracy. The integrated system achieves subpixel depth precision, dimensional errors within 0.4 to 0.95 mm, and reliable throughput on conveyor lines, offering a practical and scalable solution for smart logistics.
Zitieren
APA
Vo, C. D., Nguyen, T., & Le, Q. N. (2026). Design and Implementation of an AI-Integrated 3D Vision Conveyor System for Automated Inspection. Advanced Devices & Instrumentation, 7. https://doi.org/10.34133/adi.0201
Vancouver
Vo CD, Nguyen T, Le QN. Design and Implementation of an AI-Integrated 3D Vision Conveyor System for Automated Inspection. Advanced Devices & Instrumentation. 2026;7. doi:10.34133/adi.0201
BibTeX
@article{Vo2026,
author = {Vo, Cong Duy and Nguyen, Thanh-Truong and Le, Quang Ngoc},
title = {Design and Implementation of an AI-Integrated 3D Vision Conveyor System for Automated Inspection},
journal = {Advanced Devices & Instrumentation},
year = {2026},
volume = {7},
doi = {10.34133/adi.0201}
}
EUVIMED