基于空间邻域约束编码的视频目标跟踪研究
发布时间：2018-12-21 17:31
【摘要】：人类的伟大理想之一就是让机器人可以具备像他们自己一样的视觉功能。近一个世纪以来,信息技术飞速发展,计算机视觉方面更是科研工作者们研究的重点。到今天,计算机视觉领域的目标跟踪技术在计算精确度和跟踪实时性等方面已经达到了较高的水平。目标跟踪的目的,就是针对长度为不同时间的某些帧连续序列图像,这些序列所包含的每幅图像中均由需要被定位的运动目标。对科研工作者来说。只有符合以下标准,视频目标跟踪方法才能达标:1.实时性好,即其处理速度需要达到一定的数值;2.鲁棒性强,即面对复杂场景或目标的姿势、动作均发生大幅度改变时也不会影响算法的稳定性,仍然可以跟踪到目标。但与理论研究不同,在实际应用中,视频目标跟踪技术仍面对着诸如场景的复杂化,目标的突然变化,光照变化等多个难题。本论文针对类似对象干扰、动态模糊、低对比度、部分遮挡和光照变化等实际生活中出现在被跟踪目标所在视频序列的常见情况做出了分析研究,取得了一些主要研究成果,说明如下:1.提出一种基于空间邻域约束编码的视频跟踪方法。该方法采用了新的约束策略,即通过加权码进行双重加权的空间邻域约束编码模型,该模型是通过分别考虑特征像素的相邻像素的灰度加权编码及他们之间的欧式距离加权编码来得到的。该模型除了考虑像素本身颜色值以外,还将距离这类空间信息考虑在内,以获得在复杂场景的帧提取对应像素各种特征的健壮的代码。它进一步增强了编码的稳定性,使目标跟踪所使用的跟踪器更加健壮,在进行目标跟踪时取得了更加精确可靠的跟踪效果。2.在空间邻域约束编码的基础上,提出了其与Mean shift(均值漂移)综合后跟踪这样一种视频追踪方式。本方法在利用空间邻域约束编码模型来得到目标像素准确编码的同时,加入了Mean shift算法。Mean shift算法拥有的优势为:1.运算成本低,当待追踪标的范围确定时,能够以24帧/秒的速率进行追踪;2.即使目标产生形变,角度偏移,其边际不完全显示等情况,该算法也会排除干扰,准确追踪这两类优势。从而在确保算法健壮性的基础上提高了算法的实时性,使其面对复杂的场景时也可以做到准确定位跟踪。
[Abstract]:One of the great ideals of mankind is that robots can have the same visual function as they do. In the past century, with the rapid development of information technology, computer vision is the focus of researchers. Today, the target tracking technology in the field of computer vision has reached a high level in computing accuracy and real-time tracking. The purpose of target tracking is to target some successive sequence images of frames with different length of time. Each image contained in these sequences is made up of moving targets that need to be located. For researchers. Video target tracking can only meet the following standards: 1. Good real-time, that is, the processing speed needs to reach a certain value; 2. Robustness is strong, that is, facing the posture of complex scene or target, the stability of the algorithm will not be affected when the action changes greatly, and the target can still be tracked. However, different from the theoretical research, video target tracking technology still faces many difficulties in practical applications, such as the complexity of the scene, the sudden change of the target, the change of illumination, and so on. In this paper, we analyze and study the common situations of similar object interference, dynamic blur, low contrast, partial occlusion and illumination change, which appear in the video sequence of the target being tracked, and obtain some main research results. The description is as follows: 1. A video tracking method based on spatial neighborhood constrained coding is proposed. In this method, a new constraint strategy is adopted, that is, the spatial neighborhood constraint coding model is double weighted by weighted code. The model is obtained by taking into account the grayscale weighted coding of adjacent pixels and the Euclidean distance weighted coding between them respectively. In addition to considering the color value of pixels, the model also takes the spatial information of distance into account to obtain robust codes for extracting various features of pixels in the frames of complex scenes. It further enhances the stability of coding, makes the tracker used in target tracking more robust, and achieves a more accurate and reliable tracking effect. 2. Based on the spatial neighborhood constrained coding, this paper proposes a video tracking method which is integrated with Mean shift (mean shift. In this method, the spatial neighborhood constraint coding model is used to get the accurate coding of the target pixels, and the advantages of the Mean shift algorithm. Mean shift algorithm are as follows: 1. The operation cost is low, when the range of the target to be tracked is determined, it can be traced at the rate of 24 frames / second; 2. Even if the target produces deformation, angle deviation and incomplete display of the edge, the algorithm can eliminate the interference and track the two kinds of advantages accurately. In order to ensure the robustness of the algorithm on the basis of improving the real-time algorithm, so that it can also be accurate location tracking in the face of complex scenes.
【学位授予单位】：西安电子科技大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TN919.81
本文编号：2389230
[Abstract]:One of the great ideals of mankind is that robots can have the same visual function as they do. In the past century, with the rapid development of information technology, computer vision is the focus of researchers. Today, the target tracking technology in the field of computer vision has reached a high level in computing accuracy and real-time tracking. The purpose of target tracking is to target some successive sequence images of frames with different length of time. Each image contained in these sequences is made up of moving targets that need to be located. For researchers. Video target tracking can only meet the following standards: 1. Good real-time, that is, the processing speed needs to reach a certain value; 2. Robustness is strong, that is, facing the posture of complex scene or target, the stability of the algorithm will not be affected when the action changes greatly, and the target can still be tracked. However, different from the theoretical research, video target tracking technology still faces many difficulties in practical applications, such as the complexity of the scene, the sudden change of the target, the change of illumination, and so on. In this paper, we analyze and study the common situations of similar object interference, dynamic blur, low contrast, partial occlusion and illumination change, which appear in the video sequence of the target being tracked, and obtain some main research results. The description is as follows: 1. A video tracking method based on spatial neighborhood constrained coding is proposed. In this method, a new constraint strategy is adopted, that is, the spatial neighborhood constraint coding model is double weighted by weighted code. The model is obtained by taking into account the grayscale weighted coding of adjacent pixels and the Euclidean distance weighted coding between them respectively. In addition to considering the color value of pixels, the model also takes the spatial information of distance into account to obtain robust codes for extracting various features of pixels in the frames of complex scenes. It further enhances the stability of coding, makes the tracker used in target tracking more robust, and achieves a more accurate and reliable tracking effect. 2. Based on the spatial neighborhood constrained coding, this paper proposes a video tracking method which is integrated with Mean shift (mean shift. In this method, the spatial neighborhood constraint coding model is used to get the accurate coding of the target pixels, and the advantages of the Mean shift algorithm. Mean shift algorithm are as follows: 1. The operation cost is low, when the range of the target to be tracked is determined, it can be traced at the rate of 24 frames / second; 2. Even if the target produces deformation, angle deviation and incomplete display of the edge, the algorithm can eliminate the interference and track the two kinds of advantages accurately. In order to ensure the robustness of the algorithm on the basis of improving the real-time algorithm, so that it can also be accurate location tracking in the face of complex scenes.
【学位授予单位】：西安电子科技大学
【学位级别】：硕士
【学位授予年份】：2014
【分类号】：TN919.81
【相似文献】
相关期刊论文 前9条
1 李海洋;何东健;;一种自适应空间邻域的显著图获取方法研究[J];计算机应用研究;2013年06期
2 郑孝俊;赵海峰;罗斌;;一种医学三维数据集中感兴趣空间邻域体快速分割方法[J];计算机应用与软件;2011年09期
3 印勇;王亚飞;;基于空间邻域相关性的运动目标检测方法[J];光电工程;2009年02期
4 孟丽敏;宋余庆;朱峰;;基于空间邻域加权的模糊C-均值聚类及其应用研究[J];计算机应用研究;2010年10期
5 苏锦旗;薛惠锋;吴慧欣;;基于熵度量的空间邻域离群点查找[J];计算机工程与应用;2009年21期
6 谌德荣;孙波;陶鹏;宫久路;;基于核光谱角余弦的高光谱图像空间邻域聚类方法[J];电子学报;2008年10期
7 曾小明,谭枫;马尔可夫平稳随机域模型与遥感图象空间邻域结构信息分析[J];信息与控制;1988年04期
8 杨悦;郭树旭;任瑞治;于永力;;基于核函数及空间邻域信息的FCM图像分割新算法[J];吉林大学学报(工学版);2011年S2期
9 ;[J];;年期
相关硕士学位论文 前1条
1 赵凡迪;基于空间邻域约束编码的视频目标跟踪研究[D];西安电子科技大学;2014年
，本文编号：2389230
本文链接：https://www.wllwen.com/kejilunwen/wltx/2389230.html
