目标识别与跟踪(英文).doc_第1页
目标识别与跟踪(英文).doc_第2页
目标识别与跟踪(英文).doc_第3页
目标识别与跟踪(英文).doc_第4页
目标识别与跟踪(英文).doc_第5页
已阅读5页,还剩3页未读 继续免费阅读

下载本文档

版权说明:本文档由用户提供并上传,收益归属内容提供方,若内容存在侵权,请进行举报或认领

文档简介

Object detection and trackingAbstract: The research on moving object identification and tracking algorithms is of great theoretical value and significance. In this paper, a report is presented on the findings of moving object identification and tracking with computer vision. Object identification and tracking is a challenging problem because the tracked object moves and its size changes in the scene. Illumination changes also make it hard to identify. An approach that is able to cope with those problems using an unmanned air vehicle was applied. Unmanned air vehicle (UAV) has been an increasing field of research in both civilian and military applications. Object identification algorithmbased on HOG (Histograms of Oriented Gradients) features and SVM (Support Vector Machine) architecture. Its Continuously Adaptive Mean Shift (Camshift) algorithm that is used for object tracking.1. Background and significance of the studyVisual information accounts for a large proportion in peoples perception of environment information, and dynamic visual information is the main component of the visual information. The dynamic visual information in the perceptual environment has become an important research direction of computer vision. In real life, a lot of meaningful visual information is included in the movement. Although the human vision can see movement and can see stationary objects, people tend to more interested in moving objects in many occasions, such as aviation and military aircraft guidance, traffic monitoring, security and important places for automatic vehicle and auxiliary driving and so on.Moving object detection and tracking is a very active branch in the field of image processing and computer vision in recent years, and it is the foundation of dynamic image analysis. The target motion image sequence provides more useful information than the target at rest, so that we can use the moving object detection and tracking technology to obtain more useful information than the still image.Moving target identification and moving target tracking are of a very close relationship. As the basis of moving object tracking, moving object identification refers to a real-time detection of moving objects in the monitored scene and extracting the object. Moving object tracking is an important link in the analysis and understanding of the target detection and the upper level. The moving target tracking is based on moving object detection, using the effective characteristics of the target and the appropriate matching algorithm to find the most similar image with the template location in the image sequence. It is simple to the target location. In practical application, moving target tracking can not only provide the target trajectory, position target, provide reliable data sources for the target behavior analysis and understanding of the next step, but also can provide help for moving target detection.In summary, the research on moving target detection and tracking algorithm has great theoretical value and significance.2. Object identificationMoving object detection and tracking refers to the detection, recognition and tracking of moving objects from a sequence of images. People track the moving objects and analyze the moving objects, which can be used in safety monitoring, virtual reality, human-computer interaction and many other areas. A large number of those applications need to have reliable tracking methods that meet real-time constraints and are complex and challenging regarding changes of movement of the object, scene illumination, scale and appearance and occlusion. Tracking results can be influenced by variation of one of these parameters. It has become one of the hottest issues in the international research in recent years.For object identification, Histograms of Oriented Gradients (HOG) features and Support Vector Machine (SVM) architecture will be introduced. It uses the idea of machine learning. The method is based on evaluating well-normalized local histograms of image gradient orientations in a dense grid. Similar features have seen increasing use over the past decade. The basic idea is that local object appearance and shape can often be characterized rather well by the distribution of local intensity gradients or edge directions, even without precise knowledge of the corresponding gradient or edge positions. In practice this is implemented by dividing the image window into small spatial regions (“cells”), for each cell accumulating a local 1-D histogram of gradient directions or edge orientations over the pixels of the cell. The combined histogram entries form the representation. For better invariance to illumination, shadowing, etc., it is also useful to contrast-normalize the local responses before using them. This can be done by accumulating a measure of local histogram “energy” over somewhat larger spatial regions (“blocks”) and using the results to normalize all of the cells in the block. We will refer to the normalized descriptor blocks as HOG descriptors. Tiling the detection window with a dense grid of HOG descriptors and using the combined feature vector in a conventional SVM based window classifier gives our detection chain.Normalize gamma and colorCompute gradientsInput imageContrast normalize over overlapping spatial blocksWeighted vote into spatial and orientation cellsCollect HOGs over detection windowLinear SVMFig 1. An overview of feature extraction and object detection chain.The use of orientation histograms has many precursors, but it only reached maturity when combined with local spatial histogram and normalization in Lowes Scale Invariant Feature Transformation (SIFT) approach to wide baseline image matching, in which it provides the underlying image patch descriptor for matching scale-invariant key points. SIFT-style approaches perform remarkably well in this application. The Shape Context work studied alternative cell and block shapes, albeit initially using only edge pixel counts without the orientation histogram that makes the representation so effective. The success of these sparse feature based representations has somewhat overshadowed the power and simplicity of HOGs as dense image descriptors.Compared with other feature description methods, HOG has many advantages. First, because HOG is operated on the local square unit of the image so that it can maintain a very good invariance to image geometry and optical deformation. These two kinds of deformation will only appear in the larger space. Second, as long as the pedestrians generally be able to maintain upright posture, the algorithm allows pedestrians have some subtle body movements under the spatial sampling, coarse fine orientation sampling and strong local optical normalization condition. These subtle movements can be ignored with having no effect on the detection results. The HOG feature is particularly suitable for human detection in images. Therefore, the HOG feature is also suitable to detect the objects whose sharps cannot change.3. Object trackingIn moving target tracking, it will consume much time and computer resources if we use template matching. Because to search the whole image is necessary for template matching. However, if the targets motion state is predicted, it will greatly reduce the range of target detection, which will undoubtedly improve the efficiency of the program and also be able to improve the systems real-time performance. It will not generate false detection when some parts of the target are covered in the tracking process if the tracking method base on the color of the target. And Continuously Adaptive Mean Shift (Camshift) algorithm just base on tracking color of the target.If we apply the Mean Shift algorithm to the continuous image sequence, then we get Camshift algorithm. As the basis of the Camshift algorithm, Mean Shift algorithm is a very practical algorithm for local optimal solution. This algorithm has the characteristics of rapidity and high efficiency. It is a fast mode matching algorithm based on kernel density estimation.As a necessary step to track the object with the use of Camshift, the image acquisition from the camera and the color representation of the algorithm are not the same, which involves the issue of color space conversion. In the digital image system, there are many different kinds of color representations which are called color spaces or color models. A color space is a kind of methods to explain special status or interpretation of color characteristics in the scene. It mainly makes a simple description of the color in color domain. People use color space to simplify the color specification in some standard way. Essentially, the color space is the norm of the coordinate system and the subspace. Color space can be divided into linear color space and nonlinear color space. Commonly used color mode is a linear color space while the model is a non-linear color space. There are many possible coordinate transformations between linear space and linear space or between linear space and nonlinear space. Sometimes after space transformation, the use of another color space representation may get better results. For example, it is difficult to do color image segmentation in the RGB (red, green and blue) space due to the coupling of RGB space. However, it will be easier after being converted to HSV (hue,saturation and value) space or HIS (hue, intensity and saturation) space.Camshift algorithm is a practical algorithm to compute the local optimal solution. The algorithm has the characteristics of rapidity and high efficiency. The mean shift finds the closest model by iteration from given targets. It is a kind of method based on color information, which can track the target with certain color information. Camshift algorithm can achieve very good tracking results in some typical environmental constraints (such as a video with a simple background, including hand and arm).4. Prospectof computer visionComputer vision is a new discipline which has developed rapidly since the middle of 1960s. People use computer vision to simulate biological vision with computer and related equipment. Domestic and foreign research has made great progress. For example, computer vision plays a key role in unmanned vehicle technology. People can do 3D reconstr

温馨提示

  • 1. 本站所有资源如无特殊说明,都需要本地电脑安装OFFICE2007和PDF阅读器。图纸软件为CAD,CAXA,PROE,UG,SolidWorks等.压缩文件请下载最新的WinRAR软件解压。
  • 2. 本站的文档不包含任何第三方提供的附件图纸等,如果需要附件,请联系上传者。文件的所有权益归上传用户所有。
  • 3. 本站RAR压缩包中若带图纸,网页内容里面会有图纸预览,若没有图纸预览就没有图纸。
  • 4. 未经权益所有人同意不得将文件中的内容挪作商业或盈利用途。
  • 5. 人人文库网仅提供信息存储空间,仅对用户上传内容的表现方式做保护处理,对用户上传分享的文档内容本身不做任何修改或编辑,并不能对任何下载内容负责。
  • 6. 下载文件中如有侵权或不适当内容,请与我们联系,我们立即纠正。
  • 7. 本站不保证下载资源的准确性、安全性和完整性, 同时也不承担用户因使用这些下载资源对自己和他人造成任何形式的伤害或损失。

评论

0/150

提交评论