基于卷积神经网络的固定群体中目标人物分类
收稿日期: 2016-10-17
网络出版日期: 2017-12-30
基金资助
国家自然科学基金资助项目(61272097); 上海市教委科研创新基金资助项目(12ZZ182)
Classification of individual objects in focused group based on convolutional neural network
Received date: 2016-10-17
Online published: 2017-12-30
卷积神经网络作为深度学习的重要分支, 在图像识别、图像分类等方面有广泛的应用,其中快速特征嵌入卷积神经网络框架(convolutional architecture for fast feature embedding, Caffe) 是目前炙手可热的深度学习工具. 针对固定群体中的目标人物, 提出一种基于卷积神经网络的分类方法, 该方法不依赖于人脸图像集, 而是通过摄像头采集视频, 并利用直方图的归一化互相关方法从视频中截取训练图片, 再通过Caffe 产生训练模型, 并将个体目标图片在模型中进行匹配, 达到在固定人物群体中对个体目标进行分类的目的. 实验结果表明, 利用前期的训练模型可对固定群体中的个体目标进行准确匹配.
关键词： 固定群体; 快速特征嵌入卷积神经网络框架; 目标人物分类; 卷积神经网络
刘惠彬, 陈强, 吴飞, 赵毅 . 基于卷积神经网络的固定群体中目标人物分类[J]. 上海大学学报(自然科学版), 2017 , 23(6) : 874 . DOI: 10.12066/j.issn.1007-2861.1912
As an important branch of deep learning, convolutional neural network has been widely used in image recognition, and image classification. The convolutional architecture for fast feature embedding (Caffe) is the most popular tool in deep learning. A method of classification of individual objects in a focused group is proposed based on convolutional neural network independent of the face image set. It captures video with a camera, and obtains training images using a method of normalized cross-correlation histogram. Caffe is used to generate a training model that can realize the classification of individual objects in a focused group of people. Experimental results show that, by using a pre-training model, individual objects can be matched accurately.
[1] 明安龙, 马华东, 傅慧源. 多摄像机监控中基于贝叶斯因果网的人物角色识别[J]. 计算机学报, 2010, 33(12): 2378-2386.
[2] 宁波, 宋砚. 基于无监督方法的视频中的人物识别[J]. 计算机与现代化, 2014(12): 49-53.
[3] Zhong Y, Sullivan J, Li H B. Face attribute prediction with classification CNN [EB/OL]. (2016-02-04)[2017-09-01]. https://arxiv.org/abs/1602.01827v2.
[4] 尹萍, 赵亚丽. 视频监控中人脸识别现状与关键技术课题[J]. 警察技术, 2016(3): 77-80.
[5] 闻新, 李新, 张兴旺. 应用MATLAB 实现神经网络[M]. 北京: 国防工业出版社, 2015.
[6] Hinton G E, Salakhutdinov R R. Reducing the dimensionality of data with neural networks[J]. Science, 2006, 313(5786): 504-507.
[7] Bengio Y, Lamblin P, Popovici D, et al. Training of deep networks [M]. Massachusetts: MIT Press, 2007: 153-160.
[8] Ranzato M A, Poultney C, Chopra S, et al. Efficient learning of sparse representations with an energy-based model [C]// Advances in Neural Information Processing Systems. 2006.
[9] Krizhevsky A, Sutskever I, Hinton G E. ImageNet classification with deep convolutional neural networks [C]// Advances in Neural Information Processing Systems. 2012.
[10] Li D, Dong Y. Deep learning: methods and applications [M]. Boston: Now Publishers, 2014.
[11] Lecun Y, Bengio Y, Hinton G. Deep learning [J]. Nature, 2015, 521(7553): 436-444.
[12] Caffe [EB/OL].(2017-08-30)[2017-10-10]. http://caffe.berkeleyvision.org.
[13] Lin S H, Ji R R, Guo XW, et al. Towards convolutional neural networks compressing via global error reconstruction [C]// Preceedings of the 25th International Joint Conference on Artificial
Intelligence. 2016.
[14] Krizhevsky A, Sutskever I, Hinton G E. Imagenet classification with deep convolutional neural networks [C]// Proceedings of the 25th International Conference on Neural Information Processing Systems. 2012.
[15] He K, Zhang X, Ren S, et al. Deep residual learning for image recognition [EB/OL]. (2015-12-10)[2017-09-21]. https://arxiv.org/abs/1512.03385.
/
| 〈 |
|
〉 |
