基于对象的监控视频摘要系统的设计与实现

发布时间：2018-07-09 18:52

本文选题：视频摘要 + 目标检测　；参考：《华中科技大学》2014年硕士论文

【摘要】：随着科技的进步，人们对社会公共安全的需求不断增长，因此在很多重要场所内部署了监控摄像头，用以监控和记录场所内情况，由此产生了大量的视频数据。由于人力资源的局限性，对海量视频数据的浏览既耗时耗力，又容易遗漏重要信息。所以，如何高效浏览监控视频内容成了目前监控行业的迫切需求。通过简洁地表达监控视频数据内容，以便事后对视频的浏览、查询以及分析，成为了视频监控领域的一个研究热点和难点。针对该问题，实现了一个基于对象的监控视频摘要系统，可以对监控视频生成一个简洁且不丢失活动信息的视频摘要。主要研究内容包括：对象的检测和跟踪，对象快照的保存和索引，对象轨迹重排，摘要视频文件的生成，摘要浏览模块的实现。采用了基于样本的背景减除法实现运动对象的检测，，通过运动连续性匹配算法实现了运动对象的跟踪。对于运动对象快照的保存和索引，通过建立索引和内存映射文件技术实现了大量运动对象快照数据的快速存储和读取。采取将运动轨迹在时间轴上进行平移的策略。建立了量化时移效果的代价函数，通过模拟退火算法对代价函数进行求解，得到了时移向量。通过时移向量求出摘要帧，然后通过视频编码器将所有摘要帧合成为视频文件，从而生成了摘要视频。通过实现专门的摘要视频播放器，用户可以更方便地浏览摘要视频，获取感兴趣目标的进一步信息。实验结果表明该系统能够快速对监控视频生成摘要，大大缩短了浏览所需时间，而且保证了活动信息的完整性。
[Abstract]:With the development of science and technology, the demand for social public safety has been increasing, so a lot of video data has been generated by deploying surveillance cameras in many important places to monitor and record the situation in the places. Due to the limitation of human resources, the browsing of massive video data is time-consuming and time-consuming, and it is easy to omit important information. Therefore, how to efficiently browse video content has become the urgent need of the monitoring industry. In order to browse, query and analyze the video, it has become a hot and difficult point in the field of video surveillance. In order to solve this problem, an object based video summary system is implemented, which can generate a concise video summary without losing the active information. The main research contents include: object detection and tracking, object snapshot preservation and index, object trajectory rearrangement, summary video file generation, summary browsing module implementation. The method of background subtraction based on samples is used to detect moving objects, and the tracking of moving objects is realized by motion continuity matching algorithm. For the preservation and indexing of moving object snapshots, a large number of moving object snapshots can be stored and read quickly by establishing index and memory-mapped file technology. The strategy of moving track on time axis is adopted. The cost function of quantized time shift effect is established, and the time shift vector is obtained by solving the cost function by simulated annealing algorithm. The summary frame is obtained by time-shift vector, and then all the summary frames are synthesized into video files by video encoder, thus the summary video is generated. By implementing a special video player, users can browse the video more conveniently and obtain further information about the objects of interest. The experimental results show that the system can quickly generate the summary of the surveillance video, greatly shorten the browsing time, and ensure the integrity of the active information.
【学位授予单位】：华中科技大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TN948.6

【相似文献】