【作者】 吴伟；

【导师】 冀群心；

【作者基本信息】 吉林大学 ， 控制理论与控制工程， 2007， 硕士

【摘要】 本文出于油田防盗对图像处理的需要,对图像进行差分、检测、变换、识别,以判别图像是否为人体头肩轮廓来达到防盗的目的。本文的程序是采用Visual C++编写。考虑到油田的特殊环境会使摄像头发生抖动,所以对图像的差分要采用图像间的相互校正相减,通过与直接相减对比,提出一种运用小波的多分辨分析理论的方法来尽可能的去除背景信息并使目标图像信息损失减小。在对图像进行了差分以后,本文对图像做了一系列变换处理:先是对其进行轮廓提取,其后是归一化及图像旋转及镜像操作。经过变换以后本文对图像进行了识别,因为斜率的变化可以反映一条曲线的走向趋势,而人体轮廓曲线为一条特殊曲线。所以本文采用的识别方法为结合Haar小波与斜率的方法,通过计算其斜率的变化来判定某一目标图像是否为人体轮廓。论文给出了使用此算法的实验结果,实验结果表明,对72对144张图像的实验测试中,成功率为79.17%。更多还原

【Abstract】 The vision is the main method of the humanity gains the information from the nature . The vision must carry on processing process to the scene picture, finally achieved to the scene the understanding and the sensation, product description to the viewer. Statistics have indicated, the visual information approximately composes 60% in the information which the humanity gains, thus it can be seen the importance of visual, dynamic visual information is its main constituent, these dynamic visual information In the sensation environment have become the computer vision an important research direction.As that of image the main path of humanity won visual information, so alleged image processing,namely versus image proceed process with satisfy human vision mind or application demand. With the development of the multimedia and the computer,in recent years the repose image processing’ AMC technology at industrially get more attention than before. The former monitored control system primarily past some sensor as temperature probe, pressure gauge,in recent years,the visible control system is come out already. It was usually composed of the some cameras which are join to suite television monitor. There are always these visible system in some important location as bank, stock exchange, villa section and pantechnicon. At the same time such monitored control system can do the job,obtain video information,but it need people be there to watch the monitor,it is a very boring work for the guard to do this. In particular at some place which has many cameras, it is hard to get all things in sight. The demand for intelligent system is huge.The core of the intelligent system is the technique for extracting the information of human-body. In recent years, it becomes increasingly attractive. The technique for extraction is a composition of sensor, signal processing and pattern recognition, and has a great practicability in practice.The first step of the technique for extracting the information of human-body is separate the moving object from background. In the field of image processing, the segmentation of moving object in video sequences is a hot research topic in recent years. People often take some interests in some parts of images when they research and apply images. These special parts are often called targets and foregrounds(Other parts are called backgrounds), which correspond to the peculiar regions. Thus they can further take advantage of these targets, for example feature extraction and measurement. Image segmentation is a crucial step from image processing to image analysis, it is also a basic technique about computer vision. Because image segmentation, target separation, feature extraction and parameter measurement can change the initial images into a form of more abstract and more compact, therefore it makes possible the further analysis and understanding of image. First, this paper introduces the vision principle of moving object detection and the common methods on sequential images analyzing. Because the object is far from the camera, and also the outdoor environment has much interference, the background of gathered sequential images is often offset. Therefore, we propose a correspondence approach for feature points in sequential images. In the process of dealing with image sequences, we will revise the match if the offset of image background is big, we propose a approach which based on the idea of multiresolution analyses, the result can be improved with this methods.The recognition of moving object is the second step of the technique for extracting the information of human-body. Image recognition belongs to the pattern recognition category. Pattern recognition includes statistical pattern recognition and frame pattern recognition. The recognition of moving object has great application in practice, especially in head and shoulder detection, it is one of the important problems on human body analysis. It is the first step of face detection and the base of pedestrian detection. It can be applied to Content-based Image Indexing and Smart Surveillance Systems. It is essential to develop robust and efficient algorithms to detect head and shoulders.The contour of the people is a especial curve, so the shape description is used to identify it, shape description means to extract the information of the shape, and describe it with data structure. For the moment the shape description method is divided into two sorts, one is based on the edge, the other is based on the area. The former use the edge information of the image, the latter use the distribute information in the area. The shape description which is based on the edge is identify the edge from the image, the one which is based on the area is outstand some area which we are interested in, so we can analyse it. In this paper, we use the shape description method which is based on the edge, because the change of slope can show us the trait of a curve, we can use Haar wavelet to analyse it, and recognise whether a image is a contour of the people or not.When we know it is a contour of the people or not, we can decide to call the police or not, in this paper we are focus on the image conduct method, so we do not study the composing of Oil field guard particular.更多还原