摘要
针对现有行人跌倒检测算法在复杂场景下存在漏检、检测精度低等问题,提出一种基于注意力机制的行人跌倒检测方法YOLOX-s-EsE。改进的模型在骨干网络中引入SimAM注意力模块,在Bottleneck和特征融合模块增加ECA通道注意力模块,以进一步提取特征层的关键信息,损失函数采用EIo U,可以更有效地计算出预测框和真实框的差距,提升模型的精度。实验结果表明,改进后的算法在复杂环境下目标的检测效果有了明显的提升,相比原YOLOX-s模型,算法的mAP提高了约1.8%,达到了89.23%,精度提高了约4.6%,达到了91.79%。
Aiming at the existing problems of pedestrian fall detection algorithms such as missed detection and low detection accuracy in complex scenes,an attention mechanism-based pedestrian fall detection method of YOLOX-s-EsE is proposed.The improved model adds the SimAM attention module to the backbone network and the ECA channel attention module to the Bottleneck and feature fusion mod-ules to further extract the feature layer key information,and EIoU is adopted as the loss function,which can calculate the distance be-tween the prediction box and the ground truth more effectively and improve the accuracy of the model.The experimental results show that the improved algorithm has significantly improved the effectiveness of target detection in complex scenes.Compared with the origi-nal YOLOX-s model,the mAP of the proposed algorithm has improved by about 1.8%to 89.23%and the accuracy has improved by a-bout 4.6%to 91.79%.
作者
周蕾
钟海莲
陈冠宇
ZHOU Lei;ZHONG Hailian;CHEN Guanyu(School of Computer and Software Engineering,Huaiyin Institute of Technology,Huai’an Jiangsu 223003,China;School of Chemical Engineering,Huaiyin Institute of Technology,Huai’an Jiangsu 223003,China)
出处
《电子器件》
CAS
北大核心
2023年第2期404-413,共10页
Chinese Journal of Electron Devices
基金
江苏省教育厅自然科学基金项目(20KJA520008)
江苏省六大人才高峰项目(XYDXX-034)。
