摘要
针对行人检测中利用SSD算法不易训练、检测精度低等问题提出一种改进算法。以DenseNet作为SSD的基础网络,在其后添加四层卷积层构建新的网络;为充分利用不同深度卷积层的信息,取新建网络的后四层和DenseNet中最后两个Dense_Block来提取目标框。实验结果表明,与其它算法相比,该方法对于不同场景下行人目标检测具有更强的鲁棒性,对行人的检测率超过92%,相比改进前的算法提高10%以上。
To solve the problems of hard training and low detection accuracy in pedestrian detection using SSD algorithm,an improved algorithm was proposed.DenseNet was used as the basic network of SSD,and four convolution layers were added to construct a new network.To make full use of the information of convolution layers at different depths,the last four layers of the new network and the last two Dense_Blocks in DenseNet were selected to extract the target box.Experimental results show that the proposed method has better robustness and lower false detection and miss detection rate for pedestrian detection in different environments.The detection rate for pedestrians exceeds 92%,which is 10%higher than that of the traditional algorithm.
作者
董永昌
单玉刚
袁杰
DONG Yong-chang;SHAN Yu-gang;YUAN Jie(College of Electrical Engineering,Xinjiang University,Urumchi 830047,China;Institute of Education,Hubei University of Arts and Science,Xiangyang 441053,China)
出处
《计算机工程与设计》
北大核心
2020年第10期2921-2926,共6页
Computer Engineering and Design
基金
国家自然科学基金项目(61863033)
湖北省教育厅科学技术研究基金项目(B2016175)
湖北文理学院博士基金项目(2015B002)。
