多摄像机下的运动目标检测与跟踪-通信与信息系统专业论文.docxVIP

南京航空航天大学硕士学位论文 摘 要 多摄像机视频监控系统是现代智能视频监控系统的重要发展方向，在现代生产和生活中都 有着广泛的应用。多摄像机下的运动目标检测和跟踪是视频监控系统的重要应用功能，对这两 项技术的研究将有着重要的意义。 运动目标检测的目的是从当前视频图像中提取出可能包含目标的运动区域，并对这部分区 域进行处理，提取出我们感兴趣的目标。能否快速准确的检测出运动目标，是进行一系列后续 处理的前提和基础。 在现有的跟踪算法中，Mean shift 算法实时性好，跟踪效果佳，在目标跟踪领域得到了广 泛的应用。针对多摄像机系统中目标跟踪具有长时间、大范围场景的特点，对传统的 Mean shift 算法进行了改进，使得目标跟踪能适应视频图像中目标大小的变化，并在较为复杂的场景中也 有着良好的适应性。 相比单摄像机系统，多摄像机系统下的目标连续跟踪还需要解决两个重要问题。一个是系 统中的多摄像机协作问题，另一个是多摄像机间的目标交接问题。本文通过划分多摄像机间的 视野分界线及公共区域，结合主从摄像机分配原则完成系统对多摄像机的调度；在目标交接方 面，采用基于主颜色特征及位置匹配的方法完成对多摄像机间目标身份的确认，进而完成多摄 像机系统下的目标连续跟踪问题。 最后，对本文所做工作进行了总结和展望。 关键词：运动目标检测跟踪，多摄像机协作，目标交接，多摄像机系统，视频处理 i 多摄像机下的运动目标检测与跟踪 Abstract Multi-camera video surveillance system is an important research direction of the modern intelli- gent video surveillance system, and it is widely used in modern production and normal life. Moving target detecting and tracking with multi-cameras are important applications of video surveillance sys- tem, the research in both of them has important significance. Moving target detection is applied to abstract the moving region which may contain targets we interested from the video images. We can detect moving targets quickly and effectively by dealing with this part of the region, and this is the premise to following treatment. In the existing tracking algorithm, the Mean-shift tracking algorithm has many advantages, such as good real-time and effective target tracking, so it is widely used in target tracking field. An im- proved Mean-shift algorithm is proposed in this paper to adapt to the tracking of long time and large fields in multi-camera system. The improved algorithm can adapt to the size changes of targets in video images, and it is robust in complicated environment. Compared with single camera system, tracking moving targets continuously has two important questions to be handled. One of them is the cooperation betweens cameras in the system, the other is target handoff between multiple cameras. In this paper, we generate camera FOV lines first, then