摘要
传统的人群密度估计方法大多采用传统特征,这类特征在人群背景较为复杂时无法很好地描述人群密度情况。相关研究表明,深度特征相比传统特征能够更准确地表达图像特征信息。因此,提出了一种基于深度特征的人群密度估计方法。首先,对人群图像的原始数据进行预处理,获得训练集和测试集,分别用于分类器训练和效果检测;然后,通过卷积神经网络提取人群图像的深度特征,以此训练对应的Softmax分类器;最后,将测试集输入训练好的分类器,得到人群密度估计等级,从而实现人群密度估计。实验结果表明,使用卷积神经网络提取的深度特征能够提高人群密度估计的准确性。
Most of the traditional crowd density estimation methods use traditional features.When the background of the crowd is complex,such features cannot describe the crowd density well.Related researches show that deep features can express image feature information more accurately.Therefore,a crowd density estimation method based on deep feature is proposed in this paper.Firstly,the original data of crowd image is preprocessed to obtain the training set and the testing set,which are respectively used for classifier training and effect detecting.Then,the deep features of the crowd image are extracted by convolutional neural network and used to train the corresponding Softmax classifier.Finally,the testing set is input into the trained classifier to obtain the estimated crowd density level,so as to achieve crowd density estimation.The experimental results show that the deep features extracted by convolutional neural network can improve the accuracy of crowd density estimation.
作者
刘志
陈越
陈波
朱李楠
唐龙峰
LIU Zhi;CHEN Yue;CHEN Bo;ZHU Linan;TANG Longfeng(College of Computer Science and Technology,Zhejiang University of Technology,Hangzhou 310023,China)
出处
《浙江工业大学学报》
CAS
北大核心
2020年第3期314-318,344,共6页
Journal of Zhejiang University of Technology
基金
国家自然科学基金资助项目(61701443)
浙江省自然科学基金资助项目(LY16F020035)
浙江省教育厅项目(Y201840830)。
关键词
人群密度估计
深度特征
卷积神经网络
crowd density estimation
deep feature
convolutional neural network
