机电工程技术2025，Vol.54Issue(6)：45-50,6.DOI:10.3969/j.issn.1009-9492.2024.00151
基于全局特征下框可见劳动密集场景下工业人员检测的标签分配算法
Label Assignment Algorithm for Box-aware Worker Detection in Labor-intensive Industries Based on Global Features
摘要
Abstract
In order to solve the interference problem of personnel detection caused by complex background information and serious mutual occlusion between workers in industrial production scenarios,the key problem of label assignment in target detection algorithms is improved based on the global matching strategy in OTA.The proposed method firstly improves the selection of positive samples,and adds a weight function to the occluded and small target workers to improve the probability of being assigned.In addition,to address the problem that the remaining samples after global matching are directly assigned as negative samples,an ignore sample algorithm is proposed,which assigns anchor frames containing part of the positive sample attributes as ignored samples to avoid being directly assigned to the background class and affecting the training of the model.Finally,the experimental results are analyzed and a series of ablation experiments are conducted on the CrowdHuman dense people dataset and the self-established validation set of dense people data for industrial scenarios.The experimental results show that the model's detection accuracy for workers in dense scenes reaches 88.02%,47.91%,and 95.44%in AP,MR,and Recall metrics,respectively,which are all improved compared with traditional methods.关键词
密集人员检测/标签分配/复杂场景Key words
crowd human detection/label assignment/complex scenarios分类
计算机与自动化引用本文复制引用
何一凡,何昊阳,卢山,邵坚铭,张志铭,谢磊..基于全局特征下框可见劳动密集场景下工业人员检测的标签分配算法[J].机电工程技术,2025,54(6):45-50,6.基金项目
国家自然科学基金(62073286) （62073286）
