基于视觉仿生的视频内容分析方法研究

发布时间：2018-06-18 09:01

本文选题：视频内容分析 + 背景建模　；参考：《中国石油大学(华东)》2014年硕士论文

【摘要】：中国的视频监控市场近年来持续高速发展,国内监控摄像头正以惊人的速度逐年增加,其覆盖范围越来越广。视频信息也已成为现实生活中一个非常重要的信息载体,然而,在视频数据信息迅速膨胀的同时,视频监控系统仍停留在较低层次的简单应用上。因此,对视频监控的研究,特别是对视频内容的分析研究也越来越被关注,成为一个新的研究热点。本文以简单场景的监控视频为研究对象,针对目前视频监控存在的目标检测准确率低,目标特征提取鲁棒性差,以及视频存储周期短等问题,站在模仿人类视觉系统的角度上,采用模式识别相关算法进行研究分析,克服现有视频分析方法的局限性,以速度快、准确性高、鲁棒性好、低容量为最终目标,旨在建立一套适用于监控视频的视频处理分析系统。在深入研究了监控视频分析处理基础技术之后,确定了其分析处理的基本流程,以监控视频背景建模方法为基础进行运动目标检测,对检测出的运动目标进行稳定特征提取,并采用多特征融合的方式进行目标跟踪,最后以基于内容的背景分离处理方式进行视频编解码。其中,背景建模方法是引用了码本(CodeBook)的监控视频背景建模方法,在此基础上,本文提出了一种实时视频背景提取方法。此种背景提取技术可以实时响应背景的变化,有效提取真实背景。采用背景建模的方法检测到运动前景,分离目标图像后,通过目标帧间连续位置信息实现跟踪,对多目标帧间位置相近的情况,融合目标SIFT和彩色直方图特征进行目标匹配,并记录目标各帧的运动状态,最终实现目标运动的跟踪。实验表明该方法对多目标缓慢变化的监控视频有较好的跟踪效果。同时,本文在现有视频编解码方式基础上,提出了一种基于内容的静态监控视频编码方法。该方法是一种实现背景与运动目标分离的编码方式,采用不同的编码方式分别对背景和运动目标进行编码。实验证明,与传统编解码方法相比,基于内容的编码方法获得了更多的存储空间。论文还进一步探讨了背景更新算法对基于内容编码的影响。
[Abstract]:China's video surveillance market continues to develop rapidly in recent years, and the domestic surveillance cameras are increasing year by year at an alarming speed, and its coverage is becoming more and more extensive. Video information has also become a very important information carrier in real life. However, with the rapid expansion of video data, the video surveillance system is still in the lower level of simple applications. Therefore, the research of video surveillance, especially the analysis of video content, has become a new research hotspot. Aiming at the problems of low accuracy of target detection, poor robustness of target feature extraction, short storage period and so on, this paper takes the surveillance video of simple scene as the research object, and stands at the angle of imitating human visual system. In order to overcome the limitations of the existing video analysis methods, the research and analysis of pattern recognition algorithms are carried out. The final goal is fast speed, high accuracy, good robustness and low capacity. The purpose of this paper is to establish a video processing and analysis system for surveillance video. After deeply studying the basic technology of analysis and processing of surveillance video, the basic process of analysis and processing is determined, and the moving object detection is carried out based on the modeling method of the background of surveillance video, and the stable feature extraction of the detected moving target is carried out. Multi-feature fusion is used for target tracking and content-based background separation is used for video coding and decoding. The background modeling method is based on the CodeBook-based video background modeling method. On this basis, a real-time video background extraction method is proposed in this paper. The background extraction technique can respond to the background change in real time and extract the real background effectively. The moving foreground is detected by the method of background modeling. After separating the target image, the continuous position information between the target frames is used to track the position of the multi-target frame, and the sift and color histogram features are fused to match the target. The motion state of each frame is recorded, and the tracking of target motion is finally realized. The experimental results show that the proposed method has a good tracking effect on the surveillance video with multiple targets changing slowly. At the same time, based on the existing video coding and decoding methods, a content-based static surveillance video coding method is proposed. This method is a coding method to separate the background from the moving object. Different coding methods are used to encode the background and moving object respectively. Experimental results show that the content-based coding method obtains more storage space than the traditional codec method. The influence of background updating algorithm on content-based coding is also discussed.
【学位授予单位】：中国石油大学(华东)
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TN948.6

【相似文献】