动态特征块匹配的背景更新在运动检测的应用

发布时间：2018-11-08 08:49

【摘要】：为解决传统监控设备视场小、非智能等缺陷,结合全景成像技术和计算机视觉技术,建立无人环境下外来入侵自动检测系统,从而实现了全景监控视场下运动目标快速准确的检测及跟踪。该技术关键在于如何在复杂的动态背景下有效地提取运动目标,为此提出一种基于动态特征块匹配的自适应背景更新算法。在采用帧间差分与背景差分融合算法检测到目标的基础上,利用目标的矩信息进行跟踪,避免了全景视觉下颜色及轮廓特征缺失的弊端。根据目标的轮廓及位置提取特征块,将视频序列的每一帧图像与初始背景图像进行特征块区域的局部匹配,首先通过分析特征块图像的颜色特征,构建基于区间统计的RGB颜色直方图,提取颜色特征序列。然后通过计算序列相关性来判断该区域是否需要背景更新,从而降低对单个像素更新的冗余计算。实验表明,该更新算法具有较强的鲁棒性和可行性,能够有效提高监控系统的稳定性。
[Abstract]:In order to solve the defects of traditional monitoring equipment, such as small field of view and non-intelligent, combined with panoramic imaging technology and computer vision technology, an automatic detection system for foreign intrusion in unmanned environment was established. Thus the fast and accurate detection and tracking of moving targets under panoramic monitoring field of view are realized. The key of this technique is how to extract moving targets effectively in complex dynamic background. An adaptive background updating algorithm based on dynamic feature block matching is proposed. On the basis of detecting the target by using the fusion algorithm of inter-frame difference and background difference, the moment information of the target is used to track the target, which avoids the disadvantage of missing color and contour features under panoramic vision. According to the contour and location of the target, the feature blocks are extracted, and each frame of the video sequence is matched with the initial background image. Firstly, the color features of the feature block images are analyzed. RGB color histogram based on interval statistics is constructed to extract color feature sequence. Then, the sequence correlation is calculated to determine whether the region needs background update, so as to reduce the redundant computation of single pixel update. Experiments show that the algorithm is robust and feasible, and can effectively improve the stability of the monitoring system.
【作者单位】：长春理工大学光电工程学院;
【基金】：吉林省科技发展计划(20160520018JH) 国家自然科学基金(61605016)项目资助
【分类号】：TP391.41

【相似文献】