信息网络安全 ›› 2020, Vol. 20 ›› Issue (5): 88-93.doi: 10.3969/j.issn.1671-1122.2020.05.011

• 理论研究 • 上一篇    下一篇

视频侦查中人像智能分析应用及算法优化

张蕾华1, 黄进2, 张涛3,*(), 王生玉4   

  1. 1. 山西警察学院,太原 030401
    2. 杭州当虹科技股份有限公司,杭州 310012
    3. 公安部第三研究所,上海200031
    4. 青海省公安厅,青海西宁 810007
  • 收稿日期:2020-03-22 出版日期:2020-05-10 发布日期:2020-06-05
  • 通讯作者: 张涛 E-mail:zhangtao@stars.org.cn
  • 作者简介:张蕾华(1974—),女,山西,副教授,硕士,主要研究方向为侦查和情报等|黄进(1976—),男,福建,硕士,主要研究方向为化工机械|张涛(1982—),男,江苏,副研究员,硕士,主要研究方向为网络安全、大数据分析;|王生玉(1980—),男,青海,本科,主要研究方向为网络安全管理
  • 基金资助:
    信息网络安全公安部重点实验室开放课题(C19611)

Portrait Intelligent Analysis Application and Algorithm Optimization in Video Investigation

ZHANG Leihua1, HUANG Jin2, ZHANG Tao3,*(), WANG Shengyu4   

  1. 1. Shanxi Police College, Taiyuan 030401, China
    2. Hangzhou Danghong Technology Co., Ltd., Hangzhou 310012, China
    3. The Third Research Institute of the Ministry of Public Security, Shanghai 200031, China
    4. Department of Public Security of Qinghai Province, Xining Qinghai 810007, China
  • Received:2020-03-22 Online:2020-05-10 Published:2020-06-05
  • Contact: Tao ZHANG E-mail:zhangtao@stars.org.cn

摘要:

人像智能分析指的是对视频或录像中的人像进行结构化和可视化分析,对目标人物进行性别、年龄、发型等特征的智能识别,这项技术在视频侦查中有极高的应用价值。人像识别早期的算法是通过人工提取特征,通过学习低级视觉特征来针对不同属性进行分类学习,这种基于传统方法的模型表现常常不尽如人意。在计算机视觉领域,通过海量图像数据学习的神经网络比传统方法有更丰富的信息量和特征可以被提取。文章尝试通过深度学习技术训练神经网络模型对行人进行检测和识别,对于衣着不同的行人进行智能识别,具有更好的鲁棒性,提升了视频人像识别的准确率,拓展了人工智能技术在身份识别领域的应用。

关键词: 视频侦查, 人像识别, 深度学习, 卷积神经网络

Abstract:

Intelligent portrait analysis refers to the structural and visual analysis of portraits in video or video, and intelligent identification of the target person’s gender, age, hairstyle, etc. This technology has extremely high application value in video reconnaissance. The early algorithm of portrait recognition is to manually extract features and learn low-level visual features to classify and learn different attributes. This model based on traditional methods is often not satisfactory. In the field of computer vision, neural networks learned from massive image data have richer information and feature extraction than traditional methods. This paper attempts to train neural network models to detect and recognize pedestrians through deep learning technology, and intelligently treat pedestrians with different clothes. Recognition has better robustness, improves the accuracy of video portrait recognition, and expands the artificial intelligence technology for identity recognition.

Key words: video detection, portrait recognition, deep learning, convolutional neural network

中图分类号: