【作者】 仇晨光；

【导师】 郭书祥；

【作者基本信息】 哈尔滨工程大学 ， 控制理论与控制工程， 2010， 硕士

【摘要】 运动目标检测和跟踪技术是近年来计算机视觉、图像处理、模式识别和人工智能等领域的研究热点。该技术已经应用于智能机器人、智能交通、科学探测等领域。通常图像序列分为:静态场景图像序列和动态场景图像序列,在基于图像的目标检测和跟踪研究中,动态场景图像序列更符合实际应用、更有研究价值,因此动态场景中运动目标的检测和跟踪是该研究领域的重点。本论文主要研究了动态场景中运动目标检测和跟踪的方法。为了提高全局运动估计和补偿的精度和速度,提出了一种改进的自适应去除局部运动的方法。实现了运动目标模板的自动提取,解决了传统Mean Shift跟踪算法需要手动确定目标区域的问题。本文研究工作包括以下几个部分:首先介绍了摄像机的运动模型。在分析了摄像机在三维空域中运动的数学模型基础上,给出了目前常用的摄像机运动模型,即基于透视投影的八参数模型和基于平行投影的六参数模型。然后用基于特征的方法估计全局运动。利用SIFT算法提取具有较高精确度和稳定性的图像特征点。为了提高全局运动估计和补偿的精度,本文根据特征点运动矢量特性,提出了一种改进的自适应去除局部运动的方法。并通过实验验证了本文的算法能够取得较高估计精度和较好的补偿效果。用帧间差分法检测动态场景中的运动目标。先根据两帧图像间的全局运动参数,对两帧图像进行运动补偿,使两帧图像的背景对齐,再将补偿后的当前帧图像和参考帧图像进行差分,检测运动目标。最后利用MRF分割算法对差分的图像进行分割。采用数学形态学中闭运算的方法将目标内部的空洞和不连续的边缘填充完整,并去除孤立的噪声点,得到含有完整目标区域的二值化图像。再根据二值化图像水平投影和垂直投影的顶点坐标,实现目标区域的自动提取。解决了传统的Mean Shift跟踪算法需要手动确定目标区域的问题,最终实现了目标的自动跟踪。更多还原

【Abstract】 The moving target detecting and tracking technology has become a hot topic in the fields of computer vision, image processing, pattern recognition, artificial intelligence and so on. It has already been used in many areas such as intelligent robot, intelligent transportation and video supervising. Usually the image sequences are classified into static one and dynamic one. The latter is more tally with practice and valuable in the domain of target detection and tracking based on image sequences, so target detection and tracking in dynamic scene is the important part for this research domain.This paper mainly researches the methods of detecting and tracking of moving target in image sequences acquired by a mobile camera. In order to improve the accuracy and velocity of the global estimation and compensation, an improved adaptive noise reduction method is proposed. Target area is obtained automatically which solves the problem that the target area needs to be determined artificially in traditional Mean Shift algorithm.The primary researches of this paper include sections as following:Firstly introduce the camera motion models. The motion of camera in spatial domain is analyzed, and the typical camera motion model: the eight parameters model based on perspective projection and the six parameters model based on parallel projection are also introduced.Then the method based on feature is used for global motion estimation. The SIFT algorithm is applied for getting high accurate and stable features. In order to improve the accuracy of global estimation and compensation, it is necessary to eliminate points with local motion. An improved adaptive noise reduction method is proposed in this paper, which is effective demonstrated by the experimental results.Frame difference is used to detect moving target in dynamic scene. The motion compensation for the two frame images is done at first to get a similar background according to the estimated global motion parameters of two frame images. Then moving target is detected by the difference between the compensated image and the reference image.Finally the MRF segmentation algorithm is used for segmenting the difference images. In this paper, the holes and the edges are filled through closing operation in the field of mathematical morphology, after that, a binary image including complete target area is obtained. Then, a target template is got via mapping the coordinates of the horizontal projection and perpendicular projection vertexes to the corresponding original images. Finally, target area is picked up automatically which solves the problem that the target area needs to be determined artificially in traditional Mean Shift algorithm, and automatic target tracking is realized.更多还原