基于卷积神经网络的行人再识别算法

发布时间：2018-03-09 02:28

本文选题：行人再识别　切入点：卷积神经网络　出处：《浙江大学》2017年硕士论文　论文类型：学位论文

【摘要】：行人再识别是指输入一个摄像头拍摄的行人图像,在其他摄像头拍摄的行人图像中识别同一个人,摄像头之间没有相交的拍摄区域。随着视频监控的发展,行人再识别在智能视频监控领域的实际应用价值和理论研究意义日益凸现。研究者们提出了许多传统的解决行人再识别问题的方法,分别是从人工设计的行人特征提取和距离度量学习这两个方面出发的。在本文中,我们提出了一个基于卷积神经网络的行人再识别算法,将行人深度特征学习与距离度量学习进行融合,获得了一流的行人再识别准确率。具体而言,我们设计了专门用于行人深度特征学习的卷积神经网络模型,在最后的损失层应用了我们提出的聚类损失函数。聚类损失函数能够减小类内距离,增大类间距离,从而提高深度特征的区分度。我们在六个常用的行人再识别数据集上进行了比较实验,实验结果表明了本文提出的行人再识别算法的优势。
[Abstract]:Pedestrian re-identification refers to the input of a pedestrian image taken by a camera and the identification of the same person in the pedestrian image taken by other cameras. There is no intersecting area between the cameras. With the development of video surveillance, The practical application value and theoretical significance of pedestrian rerecognition in the field of intelligent video surveillance are becoming more and more obvious. Researchers have proposed many traditional methods to solve the problem of pedestrian rerecognition. In this paper, we propose a pedestrian recognition algorithm based on convolution neural network. The learning of pedestrian depth feature and distance measurement are combined to obtain the first class pedestrian recognition accuracy. Specifically, we design a convolutional neural network model for pedestrian depth feature learning. In the final loss layer, we apply the clustering loss function, which can reduce the intra-class distance and increase the inter-class distance. In order to improve the classification of depth features, we have made a comparative experiment on six commonly used pedestrian recognition data sets, and the experimental results show the advantages of the proposed pedestrian rerecognition algorithm.
【学位授予单位】：浙江大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.41;TP183

【相似文献】