基于自适应跟踪评价机制的视频浓缩技术的研究

发布时间：2018-05-08 12:10

本文选题：视频浓缩 + 目标检测　；参考：《山东大学》2017年硕士论文

【摘要】：随着信息化的迅猛发展以及人们对社会公共安全的日益关注,以摄像机为主要媒介的视频监控手段以其丰富、直观而具体的信息表达形式越来越得到广泛认可,获取的视频监控数据为人们在安全防范和决策支持方面都起到了举足轻重的作用。伴随着数量急剧增加的监控相机而来的是海量的视频数据,这些数据的处理需要耗费大量的人力财力物力,如何从这些数据中快速获取有价值的信息己成为行业的迫切需求。视频浓缩技术作为解决上述问题的有效方法,是对监控获取的原始视频的高度浓缩,去除大量冗余信息的同时保留视频中的关键信息,已成为监控领域的热点问题。基于对当前视频浓缩技术总体框架的研究,本文对运动目标检测、目标跟踪及轨迹提取、轨迹组合优化及图像融合这几个关键技术进行具体分析及优化。本文采用改进的视觉背景提取算法实现了运动目标的检测,完整检测出运动目标,抑制了传统算法中的"鬼影"问题;为了提高跟踪性能,解决目标丢失及浓缩视频中出现的频闪效应,本文提出一种基于目标跟踪的自适应评价机制,并对评价机制做出定性及定量分析,设计跟踪系统进行验证,进而提出基于跟踪评价机制的鲁棒跟踪算法。之后,提取和存储运动目标的完整轨迹,并建立轨迹间能量函数,将求取最优轨迹组合的问题转化为求能量函数最小值的问题,更新能量函数因子并使用模拟退火算法求取代价函数的最优解。最后根据最优轨迹组合提取出运动目标,将目标区域与背景图像融合得到浓缩视频。为了提高融合效果,消除缝合边界不自然的现象,提出阈值判断的方法,生成浏览舒适度较高的浓缩视频。通过对视频浓缩算法中各个模块进行测试,本文研发的基于自适应跟踪评价机制的视频浓缩技术能很好的去除原始视频中的冗余信息,大大缩短视频长度,节省存储空间,很好的保留原始视频中运动物体的活动信息,且不改变目标的空间一致性,真实度高,能实现用户快速浏览的需求。因此,本文提出的监控视频浓缩方法有良好的工程应用价值。
[Abstract]:With the rapid development of information technology and people's increasing attention to social public safety, video surveillance means with video camera as the main medium is more and more widely recognized for its rich, intuitive and concrete forms of information expression. The obtained video surveillance data play an important role in security prevention and decision support. With the rapid increase in the number of surveillance cameras is a huge amount of video data, these data processing needs a lot of human, financial and material resources, how to quickly obtain valuable information from these data has become an urgent need of the industry. As an effective method to solve the above problems, video concentration technology has become a hot issue in the field of monitoring, which is highly concentrated on the original video obtained by monitoring, removing a large amount of redundant information while retaining the key information in the video. Based on the research of the general frame of video concentration technology, this paper analyzes and optimizes the key technologies of moving target detection, target tracking and trajectory extraction, trajectory combination optimization and image fusion. In this paper, the improved visual background extraction algorithm is used to detect moving targets, which can completely detect the moving targets and suppress the "ghost" problem in the traditional algorithms. To solve the stroboscopic effect in target loss and concentrated video, an adaptive evaluation mechanism based on target tracking is proposed in this paper. The evaluation mechanism is qualitatively and quantitatively analyzed, and the tracking system is designed to verify it. Then a robust tracking algorithm based on tracking evaluation mechanism is proposed. After that, the complete trajectory of moving object is extracted and stored, and the energy function between trajectories is established. The problem of finding the optimal trajectory combination is transformed into the problem of finding the minimum value of the energy function. The energy function factor is updated and the optimal solution of the cost function is obtained by simulated annealing algorithm. Finally, the moving target is extracted according to the optimal trajectory combination, and then the target region is fused with the background image to obtain the condensed video. In order to improve the fusion effect and eliminate the phenomenon of unnatural stitching boundary, a threshold judgment method is proposed to generate concentrated video with high browsing comfort. By testing each module of video concentration algorithm, the video concentration technology based on adaptive tracking and evaluation mechanism developed in this paper can remove redundant information from original video, greatly shorten the length of video and save storage space. It can keep the moving information of moving objects in the original video without changing the spatial consistency of the target. It has a high degree of reality and can realize the requirement of users to browse quickly. Therefore, the video concentration method proposed in this paper has good engineering application value.
【学位授予单位】：山东大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.41

【参考文献】