【作者】 陈亮；

【导师】 陈晓竹；

【作者基本信息】 中国计量学院 ， 计算机应用技术， 2015， 硕士

【摘要】 智能视频监控系统利用计算机视觉等相关技术对视频内容进行分析处理,从而实现智能化的控制和管理。智能监控的基础是运动目标检测与跟踪技术,它们是智能视频监控系统的一个重要环节,是后续一系列分析的前提条件,其性能的好坏直接影响整个系统的表现。因此,对运动目标检测和跟踪进行研究具有重要的意思和应用价值。本文致力于复杂背景下的运动目标检测和跟踪方法的研究,研究内容主要包括以下两个方面。1、针对复杂背景环境下,鲁棒性表现突出的Vibe算法进行研究。当视频序列的第一帧中包含运动目标时,鬼影将伴随着Vibe模型的初始化而产生。由于Vibe算法并未对鬼影做任何特殊处理,鬼影的抑制完全依赖于背景像素点的穿插更新策略,导致抑制过程缓慢。为了达到快速抑制鬼影的目的,提出两种改进算法:其一,根据相邻像素点的空间一致性原则,对前景像素点增加邻域模型判断。当与邻域的匹配达到阈值时,认定该像素点为鬼影像素点,从前景中消去此前景点,并重新初始化该像素点的背景模型;其二,引入缓冲模型和第二背景模型替换Vibe算法的穿插更新策略。同时,为了避免基于单像素点判别造成的误判,对处理后的分割掩码图片进行区域检测。实验结果表明,相比于原Vibe算法,改进后的算法不仅能够有效的处理噪点和鬼影,同时能够有效处理短暂停留的前景目标。实际应用中,运动检测算法主要用于提取满足条件的运动目标。我们将改进的Vibe算法用于卡口车辆检测,发现我们提出的改进算法能够有效的检测出经过虚拟检测带的车辆。2、在现阶段的智能视频监控系统中,跟踪算法主要用于入侵检测后的目标跟踪阶段。经过对运动检测算法和跟踪算法的研究,由于Vibe算法和实时压缩感知跟踪算法在各自领域拥有良好的实时性和鲁棒性,所以我们设计了一种基于改进Vibe算法和实时压缩感知跟踪算法的区域入侵目标检测与跟踪系统。该系统能够有效的检测出入侵防区的运动目标,并对目标进行有效的跟踪。更多还原

【Abstract】 In order to achieve intelligent control and management, intelligent video surveillance system analysis and processing the video content with computer vision technology. The moving object detection and tracking, which is a key link to the following analysis and process, is not only the foundation but also a very important part of intelligent video surveillance system. Its property directly affects the performance of the whole system. Therefore, the research of moving object detection and tracking has very important meaning and application value.This paper dedicated to the research of moving object detection and tracking algorithm under complex background, the main contents include the following two parts.First, our research is based on the Vibe algorithm. As we know Vibe is a universal background subtraction algorithm, use the first frame to initialize its background model, which makes the initialization very fast. The only drawback is the moving target in the first frame will introduce an artifact called a ghost. As Vibe do nothing extra to eliminate the ghost area, so it takes a long process for ghost to fade over time. In order to solve this problem, we propose two fast eliminating ghost algorithm. First, When a pixel value was judged as foreground by Vibe, we continued to compare it with its neighboring model. If matching number reached the threshold,we identified the pixel value as ghost, then eliminated this pixel from foreground and re-initialized the pixel’s background model. Second, a new background sample won’t use to update the models of neighboring pixels; in order to avoid the misjudgment at pixel level, we detected the binary mask at blob level, which is detected by the improved Vibe algorithm; adding a second model and a cache to deal with ghost and short stay objects. The experimental results show that the improved algorithm is more effective in not only dealing with noisy point and ghost, but also short stay foreground target. Moving object detection algorithm is mainly used to extracting the moving targets which meet the conditions in reality. We find our improved algorithm is effective in detecting the vehicles passing the virtual test area.Second, now days, tracking algorithm is mainly used in target tracking stage after instruction in intelligent video surveillance system. Depends on the research ofmoving target detection and tracking algorithm, especially the vibe, we designed a tracking algorithm based on vibe and a regional invasion of target detection and tracking system which can achieve real-time compressed sensing tracking algorithm.The system can effectively detect targets invading protection zones and tracking them.更多还原