基于统一结构场模型的警务视频分析研究

项目名称： 基于统一结构场模型的警务视频分析研究

项目编号： No.61501467

项目类型： 青年科学基金项目

立项/批准年度： 2016

项目学科： 无线电电子学、电信技术

项目作者： 冯文刚

作者单位： 中国人民公安大学

项目金额： 23万元

中文摘要： 视频侦查技术是对警务视频进行分析和研判、获得有价值的线索或证据辅助侦查破案的公安技术。然而，当前公安工作中仍主要是通过人工处理完成，如能利用智能信息方法实现对于警务视频自动分析处理，则可大大提升公安部门的工作效率。本课题拟研究警务视频自动分析的统一结构场模型，以期通过构建统一结构场模型，利用共性的模型表述、状态推理和参数学习方法，尝试解决语义鸿沟问题，在警务视频发生较大变化时，仍可满足准确性、鲁棒性和效率性的组合检测需求。具体研究内容包括：1）警务视频目标提取的连续时空场模型；2）感兴趣的单语义目标检测的离散标记场模型；3）具有时序关系的目标组合检测的离散状态场模型。本项目的研究将会对于警务视频分析的理论和应用起到推进作用。

中文关键词： 警务视频；结构场模型；能量函数；参数学习推理

英文摘要： The policing video investigation technology is a policing technology to obtain valuable clues or evidences for criminal investigation, based on analyzing, studying and judging the policing videos. However, the video investigation is still by manual handling in policing routine work. If the intelligent information method could be used for automated policing video analysis, the efficiency of police combat units could be greatly enhanced. This project will build unifying structural field model to automatically analyze policing videos. This work will solve the police videos automatic analysis and bridge the semantic gap, by building unifying structural field model, which could satisfy the requirement of accuracy, robustness, and efficiency, because the model has the common model representation, status inference and parameter learning. The research of this project including three topics: 1)We build continuous spatio-temporal field model to detect the particular object on policing videos. 2) We build discrete label field to detect the specific semantic targets which is interested to the investigators. 3) We build discrete situation field model to detect semantic target bundles with sequential relationships. The project will make advantage in theory and application of policing video analysis.

英文关键词： Policing Video;Structural Field Model;Energy Function;Parameter Learning and Inference