基于改进YOLOv8的扶梯行人异常行为检测算法
Algorithm for Detecting Abnormal Pedestrian Behavior on Escalators Based on Improved YOLOv8
郭家森 1梁立振 2刘少清2
作者信息
- 1. 安徽理工大学计算机科学与工程学院,安徽淮南 232001;合肥综合性国家科学中心能源研究院,安徽 合肥2320071
- 2. 合肥综合性国家科学中心能源研究院,安徽 合肥2320071
- 折叠
摘要
扶梯行人的安全隐患多数源于乘客的异常行为.文章提出了一种基于改进YOLOv8的扶梯行人异常检测算法DEW-YOLOv8.首先为了提高算法的检测效率,在YOLOv8的C2f模块中使用效率更高DSConv模块替代标准卷积模块构成C2f_DSConv模块.其次针对行人在检测图像中所占比例较小引起的易漏检的问题,在特征融合阶段引入ECA注意力模块,弱化背景信息.最后采用WIoU损失函数替代原结构CIoU函数,提高模型的收敛能力.经过实验分析,该文的算法和原始YOLOv8算法相比,mAP@.5提升2.2％,参数量下降13.3％.
Abstract
Most of the safety hazards for escalator passengers result from abnormal behaviors of the passengers.This paper proposes an escalator passenger anomaly detection algorithm DEW-YOLOv8 based on the improved YOLOv8.Firstly,to enhance the detection efficiency of the al-gorithm,the standard convolution module in the C2f module of YOLOv8 is replaced by the more efficient DSConv module to form the C2f_DSConv module.Secondly,to address the issue of easy missed detections caused by the small proportion of passengers in the detection image,the ECA attention module is introduced in the feature fusion stage to weaken the background infor-mation.Finally,the WIoU loss function is adopted to replace the original CIoU function to im-prove the convergence ability of the model.Through experimental analysis,compared with the original YOLOv8 algorithm,the mAP@.5 of the algorithm in this paper is increased by 2.2％,and the number of parameters is reduced by 13.3％.
关键词
目标检测/YOLOv8/C2f_DSConv模块/注意力机制Key words
Object Detection/YOLOv8/C2f_DSConv Module/Attention Mechanism引用本文复制引用
出版年
2024
NETL
NSTL
万方数据