静态背景下的多目标跟踪方法研究

发布时间：2018-08-01 14:48

【摘要】：随着计算机视觉领域的飞速发展,基于静态背景下的目标跟踪方法研究成为时下计算机相关专业研究的热点。本文主要研究以视频图像为研究对象的静态背景下的多目标的跟踪方法研究,其次进行模拟图像的仿真实验。目标跟踪技术在视觉导航、智能控制、人机交互和精确制导等领域有着广泛的应用。研究静态背景图像目标跟踪方法有着其现实应用意义和价值。VIBE检测方法作为近期被提出的一种基于背景模型的目标检测方法,以其直观、快速的特点受到了国内外众多学者的关注,对VIBE检测算法进行深入的研究对目标检测技术的发展是有重要意义的。多目标跟踪领域的研究是本文研究的重点,多目标跟踪技术在现实生活具有广阔的应用前景,但其研究比较复杂,本文对常用的粒子滤波方法进行研究,在其基础上对序贯蒙特卡罗PHD算法进行研究和讨论。本文首先介绍了静态背景下多目标跟踪研究的现状和意义,其次介绍了常用的目标检测方法,并向详细介绍了 Vibe算法,并对其进行了改进。粒子滤波拥有相对其他算法可以应用于非线性、非高斯系统的能力,本文使用该方法实现了视频图像下单、多目标的跟踪问题。针对于粒子滤波粒子滤波方法应用于多目标跟踪时具有只能跟踪已知个数目标、数据关联运算复杂费时的限制,本文引入了 PHD算法,此算法可以处理未知数目目标的情况,并且PHD算法并不需要实现数据关联过程,所以这种方法的速度是优于基于粒子滤波和数据关联方法的多目标跟踪过程的,并将其应用于模拟图像仿真中。本文引入了 Science杂志中的一个简洁优秀的聚类方法-Clustering by fast search and find of density peaks,该算法对于与K-means等聚类方法相比具有无输入参数,且其简单、快捷的优点。
[Abstract]:With the rapid development of computer vision, the research of target tracking based on static background has become a hot topic in computer related research. This paper mainly studies the multi-target tracking method under the static background of video image, and then carries on the simulation experiment of the simulation image. Target tracking technology has been widely used in visual navigation, intelligent control, human-computer interaction and precision guidance. The research of static background image target tracking method has its practical significance and value. As a background model-based object detection method, VIBE is proposed recently, which is intuitionistic. Many scholars at home and abroad have paid close attention to the characteristics of fast detection. It is very important to study the VIBE detection algorithm for the development of target detection technology. The research of multi-target tracking field is the focus of this paper. The multi-target tracking technology has a broad application prospect in real life, but its research is quite complex. In this paper, the commonly used particle filter method is studied. On the basis of it, the sequential Monte Carlo PHD algorithm is studied and discussed. This paper first introduces the current situation and significance of the research on multi-target tracking in static background, then introduces the commonly used methods of target detection, and introduces the Vibe algorithm in detail, and improves it. Particle filter has the ability to be applied to nonlinear, non-Gao Si systems compared with other algorithms. In this paper, we use this method to realize the tracking problem of ordering and multi-target in video images. In view of the limitation that particle filter can only track a known number of targets when it is applied to multi-target tracking, and the data association operation is complicated and time-consuming, this paper introduces PHD algorithm, which can deal with the unknown number of targets. The PHD algorithm does not need to realize the data association process, so the speed of this method is better than that of the multi-target tracking process based on particle filter and data association method, and it is applied to the simulation of simulated images. In this paper, a simple and excellent clustering by fast search and find of density peaks, method in Science magazine is introduced. This algorithm has no input parameters compared with K-means clustering method, and it has the advantages of simplicity and rapidity.
【学位授予单位】：哈尔滨工程大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：TP391.41

【参考文献】