项目名称: 监控视频大数据中表观相似对象的判别式再标识方法与技术
项目编号: No.61471042
项目类型: 面上项目
立项/批准年度: 2015
项目学科: 无线电电子学、电信技术
项目作者: 王耀威
作者单位: 北京理工大学
项目金额: 80万元
中文摘要: 找不到有价值的信息是监控视频大数据应用所面临的一个核心技术挑战,而无约束环境下视觉表观高度相似对象间的有效标识和区分则是解决找不到问题中的难点。针对这一技术挑战,本项目围绕监控视频大数据中表观相似对象的再标识这一关键科学问题,从无约束环境下鲁棒的对象表示和高效快速的判别式对象再标识两方面开展研究,突破多模态奇异区域对象表示模型、基于深度学习的对象表示方法、多核流形学习对象再标识方法、融合监控视频编码的高效对象再标识方法等关键点。项目将搭建高清摄像机网络目标追踪原型系统,预期能实时处理16路以上高清监控视频,指定目标追踪准确率超过80%。
中文关键词: 视频监控;监控视频大数据;表观相似对象再标识;判别式学习
英文摘要: The key challenging of surveillance video big data applications is the can-not-find valuable information in massive surveillance video data. Visual appearance-like objects effective identification and recognition in unconstrained environment is the most difficult part to solve the can-not-find problem. In this project, we address this technical challenging, by focusing on the discriminative appearance-like object re-identification problem in surveillance video big data, which includes robust object representation model in unconstrained environment and effective fast discriminative objects identification. Following the key scientific problem, our project will develop new methods and techniques for multimodal singular region based objects representation model, deep learning method based objects representation method, multi-kernel manifold learning object re-identification method, and fast object identification method fusing surveillance video coding. Moreover, we will demonstrate the efficiency of the proposed model and methods on a target tracking prototype system in surveillance video. The system is designed to process surveillance video data from 8 high resolution cameras in real-time, and the tracking precision of objects in probe list is expected to over 80%.
英文关键词: Surveillance video monitoring;surveillance video big data;appearance-like objects re-identification;discriminative learning