【作者】 赵士文；

【导师】 卢珞先；

【作者基本信息】 武汉理工大学 ， 通信与信息系统， 2008， 硕士

【摘要】 图像序列中运动目标检测跟踪是计算机视觉领域一个新兴的方向和备受关注的前沿课题,它融合了计算机科学、机器视觉、图像工程、模式识别、人工智能等先进技术,广泛地应用于军事、工业、生活等各个方面。运动目标检测、背景模型建立、阴影检测、运动目标跟踪是本文研究的重点。国内外大批学者也在该领域作了深入的研究和探讨,并取得了大量的成果。本文在这些成果的基础上,在这些方面作了一定的研究和尝试,相关研究如下:(1)在运动目标检测、提取前景部分,讨论分析了目前提取前景的方法及其实现的原理。选择背景差分法作为提取运动目标的基础,为了完整、精确的提取出运动目标,采用形态学方法对差分后的前景图像进行处理;分析介绍了相关的背景建模方法,并通过试验比较建模效果。试验结果显示:高斯背景建模方法虽然能够准确的表示背景,但算法复杂度和计算复杂度方面比较复杂。针对这种情况,本文使用中值法建立背景模型和高斯背景模型相结合进行建模,其思想为:在目标区域使用高斯模型更新背景,而在非运动区域使用中值法建模。试验表明取得了很好的效果。(2)在运动目标检测跟踪过程中,很多因素导致检测跟踪出现偏差,而阴影就是其中一个主要因素。本文分析了阴影特征,阐述了目前消除阴影研究现状。针对HSI颜色模型对于当运动目标和背景颜色接近时,可能出现检测率不高的特点,把一阶梯度密度函数引入HSI颜色模型中,为了表明改进算法的有效性,利用采集到的图像序列进行实验。实验表明:基于HSI模型的阴影检测算法是有效的。(3)在运动目标跟踪部分论述了基本跟踪模型的算法原理和特点。根据运动目标在场景可能表现出的不同状态,本文给出了对相关状态的判定方法;跟踪算法使用相关匹配算法,把十字搜索法作为搜索算法。传统的搜索算法速度慢,且效率不高。为了提高搜索效率,对该算法进行了改进。实验证明可以提高搜索效率。更多还原

【Abstract】 Research on moving object detection and tracking technology in image sequences is a new arising field on the computer vision. It merges many technologies including the science of computer, machine vision, image engineering, pattern analysis, artificial intelligence, etc, is widely applied in many aspects such as military, industries, life, and so on.The paper is focus on such aspects such as moving object detection, background modeling method, shadow detecting, objects continuous tracking. Large numbers of researchers have been devoting themselves in the field and have already achieved many productions. Based on the relevant research working home and abroad, the paper introduces research and attempt. The main research can be summarized as follow:(1) The paper analyses and discusses cunent methods on retrieving foreground and the relevant principles in first part of moving object detecting and retrieving. Retrieving backing is based on the method of the background difference, and morphology can be applied to process the foreground image in order to get an integrity and true object. And then the paper introduces methods of background modeling and compares the results by the correlation experiments. According the results of the experiment, although the mixture Gaussian background model can describe background correctly, the algorithm and calculation complexity are too high. So a new background modeling method can be referred, which combines the median method background modeling with the mixture Gaussian model. The detail is: in the moving object field, Gaussian mixture background modeling can be used, and in the other field, the median method can be used. Experiments show good result can be achieved.(2) In processing of detecting and tracking moving objects, some factors lead to unexpected results, and shadow is one of the most important factors. The paper describes characteristic of shadow and the situation about how to remove shadow in home and abroad. When color in the background is similar with moving object that should be detected, it is possible for the HSI color model to bring down the efficiency of object detecting. Based the situation, the paper proposes an efficiency remove shadow algorithm which combines HSI color information model with first order gradient information model. In order to prove the efficiency, an experiment can be taken by the image sequence and analyzed the results. Experiments show improved algorithm based on the HSI color information model is efficient.(3) In part of moving object tracking, this paper introduces some basic models and algorithms principles. And then, as moving objects in the scene may present some kinds of different moving states, the paper presents the detail methods about how to judge different states by object information space. Correlation matching algorithm can be applied for tracking moving object. In the paper, correlation index can be judge whether plate matches or not, and crossing searching method is searching method. The traditional searching method is much lower and less efficiency. In order to make the searching method more efficiency, the algorithm is improved. Experiments show the result is good.更多还原