基于红外视频序列的多目标跟踪算法研究

【作者】 巩一旻；

【导师】 张毅；

【作者基本信息】 南京理工大学 ， 物理电子学， 2009， 硕士

【摘要】 运动目标识别和跟踪算法的研究是近年来图像处理与模式识别中的一个热门方向,其应用领域涉及到军事国防、天文观测、安全检测、工业生产、医疗卫生等行业。对视频中的目标进行处理,其研究对象实质是序列图像。本文在分析图像跟踪系统结构和基本方法的基础上,研究了基于灰度修正和空间滤波的图像预处理算法,为基于红外视频序列的多目标跟踪打下了较好的基础;分析了最大类间方差阈值分割法等图像分割技术以及图像匹配技术,使用最大类间方差阈值分割法与图像差分法相结合的目标检测算法,实现了目标的检测;研究了基于边缘、峰值、形心的跟踪算法,对于多目标形心点的匹配,提出了将备选形心还原到帧差图像中进行比对的方法;最后针对视频序列进行了最大类间方差法与图像差分法相结合的目标检测算法和改进后的形心跟踪算法的仿真与分析,很好地实现了基于红外视频序列的多目标跟踪,得到了较为满意的结果。本课题的创新之处在于设计了最大类间方差阈值分割法与图像差分法相结合的目标检测算法,提出了将备选形心还原到帧差图像中进行比对的方法,实现了快速有效的目标检测与真实形心点的判断。更多还原

【Abstract】 Moving target detection, recognition and tracking is the key point of the Image Process and Pattern Recognition. Its applications are related to military defense, astronomical observation, security testing, industrial production, Medical and health. To process the moving target in the video, the study is essentially of the image sequence.In this paper, the structure and basic method of image tracking system are reviewed and analyzed. On this basis, the image preprocessing algorithms based on gray-level correction and spatial filtering are studied, and lay a good foundation of multi-target tracking which is based on infrared video sequence. The image segmentation technology such as maximum variance between clusters and the image matching technology is studied. This subject adopts the new targets detection algorithm which is combined by maximum variance between clusters and image difference method to detect the targets . After analyzing the tracking algorithms of boundary tracking, peak tracking and centroid tracking, the improved centroid tracking algorithm is chosen. Then the method of comparing candidate centroids with the pixels of targets in the motion compensation prediction error information (we call it MCPR for short afterwards) images is put forward. Finally, the targets detection algorithm which is combined by maximum variance between clusters and image difference method, and the centroid tracking algorithm are simulated and analyzed for infrared video sequence, and realizes the multi-target tracking based on infrared video sequence , which obtains a favorable effect.The subject puts forward the targets detection algorithm which is combined by maximum variance between clusters and image difference method, and the method of comparing candidate centroids with the pixels in the same coordinates in MCPR image innovatively, which achieves fast and efficient targets detection and judgment of real centroid.更多还原