【作者】 王立新；

【导师】 马晓红；

【作者基本信息】 大连理工大学 ， 信号与信息处理， 2009， 硕士

【摘要】 随着计算机技术与图像处理技术的发展,运动目标的检测与跟踪逐渐成为计算机视觉领域的一个重要研究课题,并受到计算机视觉领域更加广泛的关注,其在机器人视觉导航、智能交通系统、医学图像分析等方面都有了比较广泛的应用。本文就阴影检测技术以及基于独立分量分析的运动目标检测与跟踪技术进行了深入研究分析,主要工作包括以下三部分:(1)给出了一种基于颜色不变特征的阴影检测方法。通过分析现有的运动阴影检测方法,特别是基于颜色和纹理信息的阴影检测方法,本文给出了一种基于颜色不变特征的运动阴影检测方法。该方法首先通过阴影存在判决环节来判定是否有阴影存在;然后通过预处理环节筛选出潜在的阴影像素点,在一定程度上减少了干扰的影响:再利用颜色不变特征,进一步确认各像素点是否属于阴影:最后利用阴影像素分布的几何特征,对检测结果进行修正,最终得到比较精确的阴影检测结果。(2)对基于独立分量分析的运动目标检测方法进行了改进。在对原方法进行分析的基础上,考虑到原方法存在的缺点,如分离输出顺序随机、检测精确度低、不能提取多个目标等,本文将方法中的若干环节进行了改进和完善,通过对分离输出的灰度图像进行预处理以及引入阴影检测环节充分提高了检测的精度,并通过多次交替方向的柱面投影方法,实现了多个运动目标的提取。(3)对基于带参考信号的独立分量分析的运动目标跟踪方法进行了改进。本文对带参考信号的独立分量分析方法以及运动目标跟踪问题进行了充分的分析,得出了问题产生的原因,并给出了直接以待跟踪运动目标作为参考信号的跟踪方法,即实现了目标的跟踪,降低了算法的迭代误差,又减少了方法运算量,满足了目标的快速跟踪。此外,结合因光照强度变化会导致该方法不稳定的问题,本文将归一化奇异值分解引入到该方法中,在归一化奇异值空间中,对参考信号与感兴趣输出信号估计进行相似性度量,进一步提高了该方法的鲁棒性。实验结果表明,本文给出的阴影检测方法具有良好的检测效果,对基于独立分量分析的检测与跟踪方法的改进克服了原方法的若干缺点,具有一定的理论和应用价值。更多还原

【Abstract】 With the development of the computer technology and digital image processing, moving object detection and tracking is gradually becoming a very important research subject of computer vision and is capturing more and more attention. It is widely utilized in such areas as robot vision navigation, intelligent transportation system, medical image processing, etc.In this paper, shadow detection technology together with moving object detection and tracking based on ICA are deeply researched. This paper mainly contained the following three parts:(1) A shadow detection method based on the invariant color characteristic was presented. By analyzing the present shadow detection methods, especially the methods which based on the color and texture information, this paper proposed a moving shadow detection method based on the invariant color characteristic. In this method, a step is firstly implemented to justify whether there exist shadow pixel in the current frame, then the potential shadow pixels are obtained through a preprocessing method, which reduced the disturbance to some extent; then the shadow pixels are confirmed by using the invariant color characteristic; Finally, the detection result is revised through the shadow pixels’ geometrical characteristic. As a result, the comparatively accurate detection result is obtained.(2) The moving object detection method based on independent component analysis is improved in this paper. By analyzing the original moving object detection method, especially its faults, such as the low detection accuracy, without the ability to extract multiple moving targets, this paper improved and completed the target extraction part. The detection accuracy is increased by preprocessing the gray image and introducing the shadow illumination step. The multiple targets are extracted by implementing the extracting method in which the cylinder projection is done in alternated direction for three times.(3) The moving object tracking method using independent component analysis with reference is improved in this paper. On the basis of fully analyzing the algorithm and moving object tracking, the reason why the original method, in which the normalized 2-order center moment is used as reference, generated iteration error is obtained. Consequently, a new tracking method, in which the object to be tracked is used as reference, is presented in this paper. The moving object is tracked successfully through this method and the iteration error is effectively reduced. Furthermore, considering the instability caused by the illumination change in the process of object tracking, a further solution, in which the normalized singular value decomposition is introduced to measure the similarity between the reference and the estimation of the interested output signal, is proposed. As a result, the robustness of the tracking method is improved.The experimental results demonstrated the effectiveness of the proposed methods.更多还原