基于改进AI-YOLO v4算法的施工现场安全预警技术研究

Research on construction site safety early warning technology based on AI-YOLO v4 algorithm

采用AI技术对施工现场实施安全监控,可降低人力成本和工作强度,YOLO v4是AI算法中YOLO算法的第4个版本,在目标检测中具有广泛的应用,但应用于复杂环境下存在目标漏检或重复检测的问题。为解决该问题,对YOLO v4进行改进,在YOLO v4的特定层插入注意力模块,考虑真实标定框与预测候选框的中心点距离提出新的回归损失函数,并运用多候选框训练学习策略。将改进的AI算法应用于安全帽数据集SHWD中,试验结果表明,所提出的改进YOLO v4模型对施工现场工人佩戴安全帽情况具有更高效、更精准的检测性能。

Using AI technology to implement safety monitoring on the construction site can reduce labor costs and work intensity YOLO v4 is the fourth version of the YOLO algorithm in AI algorithms,which has a wide range of applications in target detection,but it has problems of missing or duplicate detection of targets in complex environments.To solve this problem,YOLO v4 was improved by inserting an attention module into a specific layer of YOLO v4,considering the center point distance between the real calibration frame and the predicted candidate frame,a new regression loss function was proposed,and the learning strategy was trained with multiple candidate frames.The improved AI algorithm was applied to the safety helmet data set SHWD,and the test results showed that the proposed improved YOLO v4 model had more efficient and accurate detection performance for workers wearing safety helmets on construction sites.