%0 Journal Article
%T 基于YOLOv5的改进密集行人检测算法<br>An Improved Dense Pedestrian Detection Algorithm Based on YOLOv5
%A 丛笑含
%A 李士心
%A 陈范凯
%A 孟玥
%J Computer Science and Application
%P 1199-1207
%@ 2161-881X
%D 2023
%I Hans Publishing
%R 10.12677/CSA.2023.136117
%X <span style=\"font-size:14px;\">针对密集行人检测中行人之间高度遮挡重叠所带来的精度低和漏检高的问题，提出一种密集行人检测方法：Serried-YOLOv5。实验基于YOLOv5s，首先在网络特征融合阶段引入注意力机制，添加1个SE模块提高对有用信息定位的精度；然后使用Soft-NMS代替原有的NMS，保留IOU中等，但置信度较高的框，防止漏检。实验结果表明：Dense-YOLOv5相比原YOLOv5在CrowdHuman数据集上，在保证实时性的前提下，FPS提高了9.091；AP提高了1.5%；召回率Recall提高了5%，检测平均精度均值mAP0.5提升了1.5%，证明了Serried-YOLOv5方法在密集行人检测中的有效性。</span><br />
<span style=\"font-size:14px;\">In order to solve the problems of low accuracy and high missed detection caused by high occlusion overlap between pedestrians in dense pedestrian detection, a dense pedestrian detection method is proposed: Serried-YOLOv5. The experiment is based on YOLOv5s. Firstly, the attention mechanism is introduced in the network feature fusion stage, and an SE module is added to improve the accuracy of locating useful information. Then replace the original NMS with Soft-NMS, and reserve the enclosure with a medium IOU but high confidence to prevent missing checks. Experimental results show that compared with original YOLOv5 in CrowdHuman dataset, Dense-YOLOv5 has an FPS in-crease of 9.091 under the premise of ensuring real-time performance. AP increased 1.5 percent; The Recall rate is increased by 5%, and the average accuracy of mAP0.5 is increased by 1.5%, which proves the effectiveness of Serried-YOLOv5 method in dense pedestrian detection.</span>
%K YOLOv5，密集行人检测，SE，Soft-NMS，深度学习，注意力机制<br>YOLOv5
%K Dense Pedestrian Detection
%K SE
%K Soft-NMS
%K Deep Learning
%K Attention Mechanism
%U http://www.hanspub.org/journal/PaperInformation.aspx?PaperID=67284