摘要
首先采用卷积神经网络中的ALEX-NET和VGG-NET网络在Market-1501数据集上对行人特征进行提取,然后将提取出的行人特征进行相似性度量,通过计算行人准确率mAP值来实现最优匹配。实验证明,VGGNET模型较ALEX-NET模型的特征提取效果好,行人重识别的准确率高。
In this paper,firstly the ALEX-NET and VGG-NET of convolutional neural network are used to extract pedestrian features on the Market-1501 data set.Then,the extracted pedestrian features are measured for similarity,and the optimal matching is realized by calculating the mAP value of pedestrian accuracy.Experiment shows that VGG-NET model performs better than ALEX-NET model in feature extraction,and the mAP value of pedestrian recognition is higher.
作者
武忠
贺丽丽
姚利花
王峰霞
WU Zhong;HE Li-li;YAO Li-hua;WANG Feng-xia(Department of Physics and Electronic Engineering,Yuncheng University,Yuncheng Shanxi,044000)
出处
《山西大同大学学报(自然科学版)》
2019年第6期10-12,21,共4页
Journal of Shanxi Datong University(Natural Science Edition)
基金
2018年山西省高等学校教学改革创新项目[J2018169]
山西省教育科学“十三五”规划“1331工程”专项课题[ZX-18069]
关键词
行人重识别
卷积神经网络
相似性度量
mAP值
pedestrian recognition
convolutional neural network
similarity measurement
m AP value