| 题名 | 无重叠视域监控网络的轨迹挖掘与应用研究 |
| 作者 | 郝阳春 |
| 学位类别 | 工程硕士 |
| 答辩日期 | 2013-05-15 |
| 授予单位 | 中国科学院大学 |
| 授予地点 | 中国科学院自动化研究所 |
| 导师 | 台宪青 |
| 关键词 | 视频研判 标签和标注系统 多维混合数据 轨迹挖掘 Video judged tag & tagging systems high-dimensional hybrid data trajectory mining |
| 其他题名 | Trajectory Data Mining & Application For Non-Overlapping Multi-camera Network |
| 学位专业 | 计算机技术 |
| 中文摘要 | 面向公共社会安全的视觉感知数据处理，是物联网领域的重要研究方向。公安部门在进行案件侦查过程中，视频监控数据已是重要依据和线索来源。本论文直接面向案件视频研判的具体应用，收集散落在无重叠视域监控网络里的视频线索信息，用标签对嫌疑对象的特征属性进行描述，在标签数据基础上进行嫌疑对象的轨迹挖掘研究，辅助办案人员确定嫌疑对象的身份信息。具体内容如下： 首先，对案件视频数据进行收集和标注。从案发现场及附近区域的无重叠视域监控网络里，收集案件视频数据。利用面向视频图像的协同标注系统，对嫌疑对象的特征属性进行描述，以标签数据形式集中存储至线索数据库。 然后，对标签数据进行数据挖掘。因为标签数据具有高维混合类型数据的特性，按照属性类型将标签数据分为连续型、布尔型、枚举型三种子空间，分别用三种对应方法计算嫌疑对象子空间相似度，组合子空间计算加权平均值，即嫌疑对象的相似性矩阵。从相似性矩阵生成相异性矩阵，再利用凝聚式层次聚类算法，将相似度大于某阈值的嫌疑对象聚合在一起。结合视频监控摄像头的坐标位置和嫌疑对象出现的时间信息来分析嫌疑对象的运动轨迹。 最后，对比嫌疑对象的运动轨迹和在时空上重合的手机信号轨迹，确定相关手机号码来追溯嫌疑人真实身份信息。 从海量案件视频数据里发掘有价值的线索信息，辅助公安系统研判破案。作者把协同标注系统应用到案件视频数据的深度描述，以标签数据来刻画嫌疑对象。通过对标签数据挖掘聚合，描绘出嫌疑对象的运动轨迹；对比时空重合的手机号码追溯嫌疑人真实身份信息。不仅具备一定前沿理论研究价值，同时也在公安系统的视频侦查领域具备明确现实应用价值。 |
| 英文摘要 | For the public security, visual data processing is an important research direction of the Internet of Things. During the case investigation process, video surveillance data is an important basis and source of clues. Directly for the application of cases video judged, this thesis collects the video clues scattered in the non-overlapping monitoring network, use the tags to describe the characteristics of the suspects. Based on the tag data, research the trajectory of suspects, and assist police to determine the identity of the suspects. The specific contents are as follows: First, the video data of case is collected and tagged. The video data of case is collected from the non-overlapping monitoring network of crime scene and the nearby area. For the video, collaborative tagging system describes the characteristics of the suspects, centralized storage to clue database with the form of tag data. Then, the tag data is processed. Because of the tag data with high-dimensional mixed-type, in accordance with the attribute type, tag data is divided into three sub-space of continuous, boolean, enumeration type. Similarity of sub-space is calculated using three different methods, the combination of the subspace calculating the weighted average is the similarity matrix of suspects. With cohesion-hierarchical clustering algorithm, similar suspects are clustered into one group. Combined with time and space information, analyze the trajectory of the suspects. Finally, contrast the trajectory of suspects and the signal trajectory of the cell phone, determine the phone number to trace the real identity of the suspects. Discovering valuable clues from a massive video data help police judge information to solve the case. The application of collaborative tagging system is depth description of the case of video data, with the tag metadata to describe the suspects. Tag data mining shows the trajectory of the suspect. Not only has theoretical research value, and also it has clear practical value in the field of the public security. |
| 语种 | 中文 |
| 其他标识符 | 2010E8009070034 |
| 内容类型 | 学位论文 |
| 源URL | [http://ir.ia.ac.cn/handle/173211/7697]  |
| 专题 | 毕业生_硕士学位论文 |
| 推荐引用方式 GB/T 7714 | 郝阳春. 无重叠视域监控网络的轨迹挖掘与应用研究[D]. 中国科学院自动化研究所. 中国科学院大学. 2013. |
| 个性服务 |
| 查看访问统计 |
| 相关权益政策 |
| 暂无数据 |
| 收藏/分享 |
除非特别说明，本系统中所有内容都受版权保护，并保留所有权利。
修改评论