无重叠视域多摄像机监控系统中目标跟踪若干关键问题研究

发布时间：2018-02-08 18:54

本文关键词： 目标检测目标跟踪目标关联无重叠视域多摄像机　出处：《合肥工业大学》2014年硕士论文　论文类型：学位论文

【摘要】：无重叠视域多摄像机目标跟踪是指在摄像机的成像区域不重叠的前提下,实现目标在整个监控区域内的跟踪,该方向涉及运动目标的检测、运动目标静态场景下跟踪和目标关联等几个方面。在实际场景中,由于目标受到背景的干扰、目标在相机间转移存在光照和角度变化等问题,提高了多相机目标跟踪的难度。因此无重叠视域多摄像机目标跟踪系统有着实际的应用价值但同样存在极大挑战。本文在目标检测、跟踪和目标关联等相关理论方法的基础上,对监控系统的目标检测、目标跟踪和目标关联算法做了相应的研究,并实现了一个无重叠视域目标跟踪演示系统。本文主要工作成果如下：1、静态场景运动目标检测：针对自然场景光照变化和鬼影等问题,引入基于稀疏表达残差的运动目标检测算法。采用PCA的前两个主分量构建背景字典,利用字典提取重构残差,采用改进的Otsu算法获取全局阈值图像提取前景；对前景图像结合图像边缘信息剔除鬼影区域；利用保守更新的思想,采用增量PCA方法更新字典。实验表明,本文算法在目标检测方面具有很好的鲁棒性,可以有效抵抗光照的变化带来的影响。2、静态场景多目标跟踪：利用静态场景的背景不变特性,我们结合前景检测的结果,通过前景区域的匹配实现多目标跟踪。在假设前景目标已经提取出来的前提下,采用前向后向误差的思想对前景与目标进行双向最优匹配,对于没有匹配上的目标进行剔除和增加以实现静态场景下多目标跟踪。实验表明,在静态场景下,本文算法可以快速有效的跟踪上前景运动目标。3、目标关联：在假设相机间目标光线差异不大的情况下,针对目标的局部遮挡和角度不同两个问题,提出了基于稀疏表达的目标关联算法。利用分块直方图特征,将一个相机中的目标特征集加琐碎模板构建字典,对另一个相机中的目标进行重构,利用重构误差的大小来实现目标的匹配。实验结果表明,本文方法在光照变化不是非常强烈的情况下,无需先验知识,可以较准确的实现目标之间的关联。4、系统构建：将目标检测与目标跟踪相结合,构建出一个单相机下目标提取系统,将提取的前景目标通过网络传输到服务器端,利用服务器端的目标关联系统,实现目标在区域内的关联。
[Abstract]:Multi-camera target tracking without overlapped view is to realize the tracking of the target in the whole monitoring area without overlapping of the imaging area of the camera. This direction involves the detection of moving targets. In the actual scene, because the target is disturbed by the background, there are some problems such as illumination and angle change in the moving target moving between cameras. Therefore, the multi-camera tracking system has practical application value but also has great challenges. This paper based on the related theories and methods of target detection, tracking and target correlation. The target detection, target tracking and target association algorithms of the monitoring system are studied. The main work of this paper is as follows: 1. Static scene moving target detection: aiming at the problems of natural scene illumination change and ghost, etc. A moving target detection algorithm based on sparse representation residuals is introduced. The background dictionary is constructed by the first two principal components of PCA, the residual is extracted and reconstructed by the dictionary, and the foreground of global threshold image extraction is obtained by the improved Otsu algorithm. The ghost region is eliminated from foreground image combined with edge information of image, and the incremental PCA method is used to update the dictionary using conservative updating idea. Experiments show that the proposed algorithm is robust to target detection. Can effectively resist the impact of light changes. 2, static scene multi-target tracking: using the static scene background invariance, we combine the results of foreground detection, Multi-target tracking is realized by matching foreground region. Based on the premise that foreground target has been extracted, the idea of forward and backward error is used to achieve the bidirectional optimal matching between foreground and target. The targets without matching are removed and added to achieve multi-target tracking in static scenes. The experimental results show that, in static scenarios, This algorithm can quickly and effectively track the foreground moving target. 3, target association: under the assumption that there is little difference in light between cameras, the local occlusion and angle of the target are different. A sparse representation based object association algorithm is proposed. The target feature set in one camera is added to the trivial template to construct a dictionary using block histogram features to reconstruct the target in the other camera. The experimental results show that the proposed method does not require prior knowledge when the illumination variation is not very strong. The target detection and target tracking can be combined to construct a target extraction system under a single camera, and the extracted foreground target can be transmitted to the server through the network. The target association system on the server side is used to realize the target association in the region.
【学位授予单位】：合肥工业大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TP391.41;TN948.41

【相似文献】