Scholarlink

This paper investigates the possibilities offered by contemporary models of computer vision regarding object detection automation. In this work, different versions of YOLO models are studied in regard to their accuracy and efficiency corresponding to particular application conditions. Considerable effort is directed to selection of a model that is most effective for the given realistic scenarios and the model's adaptation to these tasks. Such practical issues as the application of machine vision technologies in construction, including monitoring, safety control, and process management were studied. In the research, comparative experiments were conducted on six configurations of YOLOv8 and YOLOv10 models, which used the CHV dataset with 1330 annotated images. Assessment of performance was done on selected primary metrics – mAP@50, mAP@50-95, precision, recall, and inference speed (FPS). Results suggest that YOLOv10-x surpasses all previous versions in mAP@50 with 94.2% and offers better efficiency of inference than older versions. The results obtained are fundamental in the development of optimal architectures for industrial real-time safety systems.

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

Application of YOLO Models for Automated Object Detection: Comparison, Analysis, and Recommendations

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Nurzada Amangeldy,
  • Aizhan Nazyrova,
  • Bekbolat Kurmetbek,
  • Nazerke Gazizova

摘要

This paper investigates the possibilities offered by contemporary models of computer vision regarding object detection automation. In this work, different versions of YOLO models are studied in regard to their accuracy and efficiency corresponding to particular application conditions. Considerable effort is directed to selection of a model that is most effective for the given realistic scenarios and the model's adaptation to these tasks. Such practical issues as the application of machine vision technologies in construction, including monitoring, safety control, and process management were studied. In the research, comparative experiments were conducted on six configurations of YOLOv8 and YOLOv10 models, which used the CHV dataset with 1330 annotated images. Assessment of performance was done on selected primary metrics – mAP@50, mAP@50-95, precision, recall, and inference speed (FPS). Results suggest that YOLOv10-x surpasses all previous versions in mAP@50 with 94.2% and offers better efficiency of inference than older versions. The results obtained are fundamental in the development of optimal architectures for industrial real-time safety systems.