【作者】 顾梅；

【导师】 徐涛；

【作者基本信息】 南京航空航天大学 ， 计算机应用技术， 2007， 硕士

【摘要】 近年来,彩色视频序列中目标跟踪技术的研究方兴未艾,运动目标跟踪已经成为模式识别、图像处理、计算机视觉、武器制导等研究领域的重要课题,在军事、交通、生物医学等多种领域都具有重要的意义,发展应用前景广阔。主动形状模型(ASM)方法通常是利用图像中的灰度信息对非刚性物体进行定位跟踪,由于利用了先验知识,因此在目标定位跟踪中比其他方法具有更好的鲁棒性。本文将图像中的颜色信息结合ASM方法扩展应用到视频序列中,提出了一种彩色视频序列中运动目标的定位跟踪方法。首先使用人机交互方式提取目标轮廓的边界点,并形成训练样本集;对训练样本集中的所有形状向量进行对齐操作后,运用主分量分析(PCA)方法建立目标轮廓的先验模型,先验模型的参数反映了形状的主要可变化模式。利用先验模型搜索运动目标时首先通过局部颜色特征模型匹配得到各个边界点的更佳的位置,经过相似性变换调整模型参数后,再进行局部颜色特征模型的匹配,形成一个迭代过程,模型与目标物体实际轮廓逐步逼近,从而对目标准确匹配达到目标定位跟踪的目的。本文以彩色视频序列中运动的蝴蝶鱼为研究对象进行一系列实验,通过对实验结果进行分析和比较可以发现,在该方法中使用颜色信息能有效地实现彩色视频序列中的目标定位跟踪,同时利用视频中相邻帧间的相关信息确定模型初始位置,有效提高了目标定位跟踪的效率。更多还原

【Abstract】 For the past few years, the research on object tracking technique in color video sequences is in the ascendant. Moving object tracking has become an important research topic in many study fields, such as pattern recognition, image processing, computer vision and weapon control-guide, etc. It plays very important roles in military, traffic and biomedicine, etc, and it is of a wide development prospect as well.By making use of gray information in images, active shape model (ASM) usually performs non-rigid object localization and tracking. The model shows its better robustness compared to other localization and tracking methods because of the use of the prior knowledge. Combining color information in images with active shape model and extending the improved model to video sequences, this paper presents an algorithm for moving object localization and tracking in color video sequences. First landmark points are extracted for object silhouette by human-computer interactive mode, then the landmark points form into training sample set. Aligning all the shape vectors in the training sample set, the prior contour model is constructed based on principal component analysis (PCA) technique. The parameters of the prior contour model show the mostly variable mode of the shape. First the prior contour model is put initially in the image, and then the shape model and local color feature model are needed to be trained alternately. The shape model is iteratively changed until it converges to the object silhouette that a good overall match is obtained in a few iterations.Finally through comparison of test results based on the moving butterfly fish in color video sequences, the method presented in this paper combining with the color information shows effective results for object localization and tracking in color video sequences. And the determination method of initial position presented improves object localization and tracking efficiency remarkably.更多还原