多摄像机无重叠视野域的目标跟踪

发布时间：2018-04-14 14:15

本文选题：多摄像机 + 无重叠视野域　；参考：《北京交通大学》2014年硕士论文

【摘要】：随着社会的发展,智能监控技术逐步成为计算机视觉领域研究的重点。为了满足监控场景的需求,全面的获得运动目标信息,需要多个摄像机的协同合作。而对监控范围和计算量等因素的综合考量,无重叠视野域的多摄像机目标跟踪技术受到了广泛的关注。在多摄像机无重叠视野域的监控环境中,由于不同摄像机视野域内的光照情况、目标姿态等因素的不同,造成同一目标在不同摄像机中的成像有所不同,进而给多摄像机之间的目标匹配带来困难；同时,由于多个摄像机之间存在盲区,当运动目标消失在盲区时,系统无法获知目标的运动信息,进而无法实现对目标的连续跟踪。针对以上问题,本文对无重叠视野域多摄像机的目标跟踪技术进行了研究,主要工作如下：第一,实现了单摄像机下的目标检测与跟踪,并对阴影去除和目标跟踪算法做了一定的改进。第二,针对多摄像机无重叠视野域的目标跟踪匹配问题,本文提出了两种目标匹配方法：(1)在有训练集的情况下,提出了一种基于亮度转换函数子空间的分层目标匹配法,利用概率主成分分析法对一对摄像机之间的亮度转换函数空间进行降维处理,并采取双向映射,再结合目标颜色分布的区域特征对目标进行匹配；(2)没有训练集的情况下,在量化的HSV空间中,利用信息熵描述目标颜色分布对目标颜色特征的贡献程度,在目标颜色特征中融入颜色空间分布信息,从而提高了目标匹配的准确度。第三,提出一种基于Agent的多摄像机无重叠视野域的目标跟踪方法。利用智能Agent代理摄像机,从而使摄像机具有Agent的特性。多个摄像机之间可以通过消息通信进行协同合作,进而对监控场景中出现的遮挡、盲区等现象给目标跟踪带来的问题进行处理。第四,利用JADE平台搭建多Agent系统,初步实现多摄像机智能代理系统。并利用真实视频数据验证基于Agent的多摄像机无重叠视野域的目标跟踪方法的有效性。
[Abstract]:With the development of society, intelligent monitoring technology has gradually become the focus of computer vision research.In order to meet the needs of the monitoring scene and obtain the moving target information comprehensively, the cooperation of multiple cameras is needed.However, considering the monitoring range and computational complexity, the multi-camera tracking technology without overlapping field of vision has been paid more and more attention.In the surveillance environment of multi-camera with no overlapping field of vision, the imaging of the same target in different cameras is different due to the different illumination and attitude of the target in different camera field.At the same time, because of the blind area between multiple cameras, the system can not get the moving information of the target when the moving object disappears in the blind area, and then can not realize the continuous tracking of the target.Aiming at the above problems, this paper studies the target tracking technology of multi-camera with no overlapping field of view. The main work is as follows:First, the target detection and tracking under single camera is realized, and the shadow removal and target tracking algorithms are improved.Secondly, aiming at the problem of target tracking and matching without overlapping field of view of multiple cameras, this paper proposes two target matching methods: 1) in the case of training set, a hierarchical target matching method based on brightness conversion function subspace is proposed.Using probabilistic principal component analysis (PPCA), the brightness conversion function space between a pair of cameras is reduced, and bidirectional mapping is adopted to match the target with the regional features of the color distribution.In the quantized HSV space, the information entropy is used to describe the contribution of the object color distribution to the target color feature, and the color space distribution information is incorporated into the target color feature, thus the accuracy of target matching is improved.Thirdly, a target tracking method based on Agent is proposed.The intelligent Agent proxy camera is used to make the camera have the characteristics of Agent.Multiple cameras can cooperate with each other through message communication to deal with the problems of target tracking caused by the occlusion and blind area in the monitoring scene.Fourthly, the multi-Agent system is built on JADE platform, and the multi-camera intelligent agent system is preliminarily realized.The real video data is used to verify the effectiveness of the target tracking method based on Agent.
【学位授予单位】：北京交通大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TP391.41;TN948.41

【参考文献】