【作者】 杨军；

【导师】 林俊；

【作者基本信息】 湖南师范大学 ， 电路与系统， 2015， 硕士

【摘要】 随着社会经济的快速发展,计算机视觉技术也越来越成熟。视频监控系统的应用更加广泛,智能视频监控技术的发展可以方便人的生活,减少人的参与。智能视频监控技术在交通管理中的地位也越来越高,起到节约人力资源,提高科学化管理水平,减少安全事故的作用。可见视频监控技术是计算机视觉领域的热点与难点。论文主要研究内容:统计监控视频中两个方向行驶的船舶数量和监控画面中船舶的数量,统计渡口监控视频中等待渡船的人数以及相关图像处理算法,对比分析了几种常见的目标检测算法。前景目标的检测是准确统计的前提和基础,针对河段监控视频背景固定的特点,本文采用背景差分法来检测图像中的船舶以及人体目标。船数的检测主要是对场景中运动目标的检测,而人的检测既包括场景中的运动目标也要包括静止目标,由于两种检测统计算法检测的对象不同,使用了两种不同的背景更新方法。对船数的统计是使用全局迭代的方法来更新背景,人数统计中用背景相对固定的差分法来检测运动缓慢或静止不动的目标。当场景中无统计目标时,背景完成更新,以适应外界环境的变化。船数的统计中采用一种基于虚拟计数线的方法来实现船舶数量的统计。算法首先在监控区域中利用OpenCv里面的ROI来设置感兴趣区域。然后根据目标与计数线位置关系进行计数,即当进入ROI区域的对象满足设定的条件时,距离其最短的计数线进行计数。该方法的计数准确率较高,但是船体的相互遮挡和重叠导致误差的存在。在人数统计中,本文首先完成对前景目标的检测,然后根据目标在视频画面中的位置与前景目标区域面积的关系来划分区域,分别检测区域中对象的轮廓尺寸是否满足本区域设定的条件,如果满足条件,标记其轮廓并且系统进行计数。本人数统计算法在目标发生重叠遮挡时存在误差。论文以提高算法的准确性和实用性为目标,实现了对视频监控河段的船舶数量统计和渡口乘客的数量统计,取得了较好的效果,并对试验截图进行分析。更多还原

【Abstract】 With the rapid development of social economy, the computer vision technology is more and m ore mature. Application of video m onitoring system was more widely, the d evelopment of intelligent video surveillance technology could facilitate people’s life, the participation of the people was reduced. Status of intelligent video surveillance technology in traffic management is also more and m ore broad to save human resource, improving the scientific management level, reducing the role of safety accident. Visible video monitoring technology is a hot and difficult points in the computer vision field.The main research is about counting the num ber of ships both passing in different directions and in the surveillance video, counting the number of ferry passengers in the su rveillance video and discussed th e related image processing algori thms. Foreground object detection is the premise and basis of accurate sta tistics, as the surveillance video background is fixed, the paper used the background subtraction to detect ships and body tar get. Detecting the sh ips is mainly to detect moving objects in the scene, The human detection including not only m oving objects in the scene but also the stati onary target. As this two statistical algorithms detecting dif ferent objects, two dif ferent methods of background updating methods were used in the pape r. The statistics of ships taking global iteration to update the background, the statistics of people using relatively fixed background to detect slow or static target. If there is no statistical tar get in the background, the background was updated to adapt to the changing environment.A statistical method based on statis tical counting lines was used to achieve the nu mber of ships statistic s. Firstly, the ROI inside Open Cv was used to set the region of intere st, then according to the positional relationship between the target and the counting line to complete counting. When the target entered in the m onitored area and marked, the counting line shortest to the tar get counted. Accuracy of the method is well. However the hull mutually overlap le ading to occlusion of errors. The paper complete the detec tion of foreground object s in statistics of ferry passengers firstly, then based on the re lationship of the target area in the video screen and its regional area to zoning, When detectin g the area connected domain meeting in the region of the target setting in t he conditions, marked its ou tline. There is error of statistical algorithm s when the shelter happened in the target area.The statistics of the ships and fe rry passengers were achieved based on the accuracy an d practicality of the algorithm, th e results was satisfactory and the test shots were analyzed.更多还原