基于改进YOLOv5s的车载人员安全带行为检测

车载人员佩戴安全带行为的检测对于人的生命安全保障具有重要作用。针对目前车内复杂环境下车载人员佩戴安全带检测精度不高的问题，提出一种基于改进的YOLOv5s(You Only Look Once v5s)车载人员佩戴安全带的检测方法。该检测方法将YOLOv5s作为基础网络，在此基础上进行改进。为改善深度模型对特征信息的提取能力，采用RFB(Receptive Field Block)模块增大网络的感受野，并利用RFB模块多分支结构获得混合的感受野；加入ECA(Efficient Channel Attention)注意力通道模块，使得整个网络更加专注特征信息的提取；将原YOLOv5s的损失函数替换为EIOU，进一步提高网络对安全带的检测精度。经过实验结果表面，改进后网络与原YOLOv5s网络相比，其平均精度均值(mAP，mean Average Precision)提高了2.2%，查准率(Precision)提升了5.1%。改进后的网络具有良好的提升效果，表明了该方法的有效性。

Seatbelt behavior detection of vehicle occupants based on improved YOLOv5s

The detection of seatbelt wearing behavior of vehicle-borne personnel plays an important role in ensuring human life safety. Aiming at the problem of low detection accuracy of seat belt worn by vehicle occupants in complex environment, an improved detection method based on YOLOv5s(You Only Look Once v5s) was proposed. The detection method takes YOLOv5s as the basic network and improves on it. In order to improve the ability of the depth model to extract feature information, the Receptive Field of the network is expanded by using the receptive field RFB(Receptive Field Block) module, and the hybrid receptive field is obtained by using the multi-branch structure of the RFB module. Adding the Efficient Channel Attention ECA(Efficient Channel Attention) module makes the entire network focus more on extracting feature information. The loss function of the original YOLOv5s is replaced by EIOU to further improve the detection accuracy of the safety belt. The experimental results show that compared with the original YOLOv5s network, the mAP (mean Average Precision) of the improved network is increased by 2.2%, and the Precision is increased by 5.1%. The improved network has good enhancement effect, which shows the effectiveness of the method.