摘要
为了提高行人检测的准确率和检测速度,提出一种基于全卷积单目标检测算法(FCOS)改进的轻量级行人头部检测算法(HDNet).算法使用M-ShuffleNetV2网络作为模型骨干网络,提高模型的检测速度;在局部空间特征融合模块中引入特征金字塔结构,通过对不同特征层之间的特征进行融合,增强模型的局部特征提取能力;在全局混合特征注意力模块中引入注意力机制,增强模型对全局特征的提取和使用能力.实验结果表明:算法在公开数据集BrainWash、Scut-Head-A和Scut-Head-B上的平均精度(AP)分别为92.0%、91.9%和92.3%,优于其他相关算法;算法的每秒传输帧数(FPS)为54.0,优于YOLOv3算法的52.1和FCOS算法的28.1.改进后算法能够充分利用行人的有效特征,满足了复杂场景下行人头部检测任务的各项要求.
To improve the accuracy and detection speed of pedestrian detection,an improved lightweight pedestrian head detection network(HDNet)based on fully convolutional one-stage object detection algorithm(FCOS)is proposed.The M-ShuffleNetV2 network is used as a backbone network to improve the detection speed of the model.The feature pyramid network is introduced into the local spatial feature fusion module,and the local feature extraction ability of the model is enhanced by fusion of features between different feature layers.The attention mechanism is introduced into the global hybrid feature attention module,which enhances the model’s ability to extract and use global features.Experimental results show that the average precision(AP)of the algorithm on the public data sets BrainWash,Scut-Head-A and Scut-Head-B are 92.0%,91.9%and 92.3%,which is better than other related algorithms.The frames per second(FPS)of the algorithm is 54.0,which is better than 52.1 of the YOLOv3 algorithm and 28.1 of the FCOS algorithm.The improved algorithm can make full use of the effective characteristics of pedestrians,thus meeting the requirements of pedestrian head detection tasks in complex scenes.
作者
林文杰
邵家玉
张宁
Lin Wenjie;Shao Jiayu;Zhang Ning(School of Automation,Southeast University,Nanjing 210096,China;Key Laboratory of Measurement and Control of Complex Systems of Engineering of Ministry of Education,Southeast University,Nanjing 210096,China;ITS Research Center of Ministry of Education,Southeast University,Nanjing 211189,China)
出处
《东南大学学报（自然科学版）》
EI
CAS
CSCD
北大核心
2022年第6期1152-1160,共9页
Journal of Southeast University：Natural Science Edition
作者简介
林文杰(1998—),男,硕士生;联系人:邵家玉,男,博士,副教授,101004809@seu.edu.cn.
