【作者】 张林；

【导师】 魏祖宽；

【作者基本信息】 电子科技大学 ， 计算机软件与理论， 2014， 硕士

【摘要】 视频监控是安全防范系统的重要组成部分,它以其直观、准确、及时和信息内容丰富而被应用于许多场合。智能视频监控是对传统监控系统的智能化,从而减轻了人员的压力,扩大了监控范围。运动目标跟踪是智能视频监控的主要技术之一,而单摄像机跟踪系统的视野域有限,跟踪目标容易受到其他障碍物的干扰,因此研究多摄像机环境下的运动目标跟踪具有非常重要的意义。本文以变电站场景下的运动目标跟踪为应用背景,研究了重叠多摄像机的目标跟踪方法,主要包括以下内容:1.改进了一种基于一维标定物的多摄像机标定算法。该算法根据已知的点对应得到每组摄像机的基础矩阵,然后利用每个基础矩阵得到对应的每组摄像机的射影矩阵。为了从射影矩阵得到欧氏矩阵,完成多摄像机的标定,可以经过如下两个阶段得到由射影矩阵转化为欧氏矩阵的变换矩阵:第一阶段是求解第一组摄像机的变换矩阵,第二阶段是使用不同方法得到其他组摄像机的变换矩阵。2.为了给运动目标跟踪算法提供输入数据,首先比较了背景减除法、一维高斯模型、高斯混合模型和核密度估计四种基于背景的前景目标提取算法,综合考虑之后选择背景减除法作为运动目标提取算法。然后详细叙述了概率占用图的构建方法,该概率占用图以运动目标的前景二值图像作为输入,产生目标在场景中出现的概率的近似值,该近似值可以作为后续运动目标跟踪算法的输入数据。3.提出了结合k最短路径和基于颜色的线性规划算法来求解多目标运动轨迹的方法。首先根据多个摄像机的监控区域建立网络流,以目标在监控区域中出现的后验概率计算各个时刻目标最有可能经过的网络节点。为了加快求解速度,提出了利用k最短路径算法作用于网络流以得到k个目标运动轨迹的方法。但是k最短路径跟踪算法没有考虑目标的颜色信息,因此接下来提出了使用k最短路径与基于颜色的线性规划相结合的方法,这种方法不仅降低了计算的时间复杂度,而且利用了目标的颜色信息,提高了跟踪的准确性。4.设计了一个变电站场景下的跟踪系统。该系统使用多个摄像机对变电站的室内室外区域进行实时监控,利用本文提出的多摄像机跟踪算法对监控区域中的运动目标进行实时跟踪,同时保存监控数据,为事后管理人员提供帮助。更多还原

【Abstract】 Video surveillance which is used in many occasions is an important part of the security system, with its intuitive, accurate, timely and rich information content. Intelligent video monitoring is for intelligent of traditional monitoring system, so as to reduce the pressure of the personnel and expand the scope of monitoring. Moving object tracking is one of the major technical of intelligent video surveillance. As single-camera tracking system has a limited field of vision and his tracking objects are vulnerable to the interference of other obstacles, the research of moving object tracking in multi-camera environment has very important significance.This paper based on the application of moving object tracking in substation scenarios has studied the object tracking method of overlapping multi-camera, and mainly includes the following contents:1. Improved a multi-camera calibration algorithm based on one-dimensional calibration pattern. The algorithm is based on the known point correspondences to get the fundamental matrix of each set of cameras, and then utilizes each fundamental matrix to obtain projection matrices of corresponding set of the cameras. To get Euclidean matrices from projection matrices and complete multi-camera calibration, the following two phases to obtain transformation matrices which make projection matrices transform into Euclidean matrices could perform: the first phase is to solve the transformation matrices of the first set of cameras, and the second phase is using different methods to obtain the transformation matrices of other set of cameras.2. To provide input data for moving object tracking algorithm, the second chapter describes two aspects. Firstly, compared four algorithms of foreground object extraction based on background including Background Subtraction Method, One Gaussian Model, Gaussian Mixture Model and Kernel Density Estimation, and selected Background Subtraction Method as moving object extraction algorithm after the comprehensive consideration. Secondly, described in detail the construction method of probabilistic occupancy map which using foreground binary image of moving object as input data, producing the approximation of probability of the object in the scene. The approximation can be used as input data of the following moving object tracking algorithm.3. Proposed a method of k-shortest paths combining linear programming algorithm based on color information to solve multi-objective trajectory. Firstly, established a network flow based on monitoring area of multi-camera, and computed the network nodes that each time the object is most likely to pass, based on the posterior probability of object in the monitoring area. Secondly, to speed up the computation, proposed a method which uses k-shortest paths algorithm acting on network flow to get k object moving trajectory. However, k-shortest paths algorithm has not consider the color information of objects. So proposed a method of k-shortest paths combining linear programming algorithm based on color information to reduce the computational time complexity and also use the color information of object to improve the tracking accuracy.4. Designed a tracking system in substation scenarios. The system uses multi-camera to achieve real-time monitoring in indoor and outdoor areas of substation, and utilizes multi-camera tracking algorithm proposed in this paper to realize real-time tracking moving objects in monitoring area, at the same time, preserves the monitoring data to help managers.更多还原