视频中运动目标检测与追踪算法研究

发布时间：2018-03-27 18:57

本文选题：视频监控技术　切入点：目标检测　出处：《安徽工程大学》2017年硕士论文

【摘要】：智能视频监控技术是一门将图像处理技术、模式识别技术、电子传感器技术、计算机数据挖掘分析技术相结合的综合性技术。飞速发展的社会经济,给智能视频监控技术在智能安防、智能交通、智能旅游、城市管理与规划方面提供了广阔的应用前景。因此,开展智能视频监控技术理论与方法的研究,具有重大的现实意义和科学价值。就智能视频监控技术而言,运动目标检测与目标追踪是两个主要研究的方面,是保证视频监控系统获得准确、有效信息的重要前提。目前,这方面的研究引起了国内外众多专家与学者的广泛关注,成为当下的研究热点。本文根据目前智能视频的发展趋势,针对目标检测与目标追踪的理论与算法展开一定的工作与研究,论文主要完成工作与创新点:1.关于目标检测方面的内容,本文主要分析了 SIFT、SURF算法在特征点提取、描述子生成、特征匹配方面的具体过程,并通过MATLAB仿真软件模拟在光照、扭曲变形、噪声干扰情况下,两种特征点匹配算法的匹配效果。2.阐述了压缩感知理论知识,通过稀疏信号表示、测量矩阵选择、信号重构算法三个方面内容进行分析。3.研究一种融合SURF特征的压缩追踪算法,该算法在压缩追踪算法的基础上进行改进。采用SURF算法求得前后两帧中目标特征点的匹配关系,求解追踪目标尺寸变化,自适应调整目标模板大小。利用压缩追踪算法对目标进行压缩后追踪,在压缩追踪算法自适应更新目标外观模型的基础上,增加误更新外观模型判断机制,解决追踪过程中严重遮挡和扭曲变形问题。4.研究一种鲁棒的时空上下文算法,该算法针对时空上下文在遮挡情况下的不足进行优化。在目标检测阶段,采用SIFT算法对目标进行初步定位,减少前期搜索的时间和搜索范围。在目标追踪阶段,采用时空上下文算法的思想,利用目标周围的时空信息辅助追踪。当遭遇严重遮挡的情况下,不再采用时空上下文算法,改用卡尔曼滤波算法对目标进行预测。
[Abstract]:Intelligent video surveillance technology is a comprehensive technology which combines image processing technology, pattern recognition technology, electronic sensor technology and computer data mining technology. It provides a broad application prospect for intelligent video surveillance technology in intelligent security, intelligent transportation, intelligent tourism, urban management and planning. As far as intelligent video surveillance technology is concerned, moving target detection and target tracking are two main aspects of research, which is an important prerequisite to ensure the video surveillance system to obtain accurate and effective information. The research in this field has attracted the extensive attention of many experts and scholars at home and abroad, and has become the current research hotspot. Aiming at the theory and algorithm of target detection and target tracking, the thesis mainly completes the work and the innovation point: 1. About the content of target detection, this paper mainly analyzes the SIFT SURF algorithm in feature point extraction and descriptor generation. This paper simulates the matching effect of two feature point matching algorithms in the case of illumination, distortion and noise interference by MATLAB simulation software. 2. The theory of compressed perception is described, and the sparse signal representation is presented. The measurement matrix selection and signal reconstruction algorithm are analyzed. 3. A compression tracking algorithm based on SURF features is studied. The algorithm is improved on the basis of compression tracking algorithm. SURF algorithm is used to obtain the matching relation between the target feature points in the two frames before and after, and the change of the target size is solved. The target template size is adjusted adaptively. The compression tracking algorithm is used to track the target after compression. On the basis of the compression tracking algorithm to update the object appearance model adaptively, the judgment mechanism of the false update appearance model is added. To solve the problem of severe occlusion and distortion in tracking process. 4. A robust spatio-temporal context algorithm is studied, which is optimized for the lack of occlusion in spatio-temporal context. The SIFT algorithm is used to locate the target in order to reduce the time and scope of the previous search. In the phase of target tracking, the idea of spatio-temporal context algorithm is used. In the case of severe occlusion, the spatio-temporal context algorithm is no longer used and the Kalman filter algorithm is used to predict the target.
【学位授予单位】：安徽工程大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.41

【参考文献】