【作者】 吕林；

【导师】 吴巍；

【作者基本信息】 武汉理工大学 ， 通信与信息系统， 2012， 硕士

【摘要】 目标跟踪技术是现代图像处理以及视觉研究领域内最热门的研究课题之一。随着科技的发展,特别是计算能力以及通信技术的飞速发展、图像识别领域算法的突破性改进,为目标跟踪技术的出现打下了坚实的理论基础和实现的可能,近年来,目标跟踪技术在天文预测、军事国防、智能监控、交通监视等领域有着及其重要的实用价值。本文介绍了海天背景下舰船目标跟踪的国内外研究现状、传统的目标跟踪算法的技术与相关理论,主要的工作是将粒子滤波算法应用到海天目标跟踪领域,并且针对当前的跟踪理论提出了适用于海天目标跟踪的粒子滤波跟踪算法,为开发实时高效的海天背景目标跟踪系统做出前期研究,具体的研究内容是：(1)本文总结和研究了几种传统的目标跟踪理论,介绍了基于卡尔曼滤波算法和Mean-shift算法的滤波原理,针对这两种算法分别进行了一段视频跟踪实验仿真,并在卡尔曼滤波算法下的海天目标跟踪进行了误差仿真,比较了Mean-shift算法下,用巴氏系数比较搜索到的目标与初始帧目标的匹配程度。(2)本文总结了海天背景下传统粒子滤波器的跟踪算法原理,针对目前在目标跟踪中进行特征选取的单一性,提出综合利用包含目标的颜色直方图和纹理直方图的观测信息,将目标的颜色和纹理融合之后作为目标的跟踪特征,建立一种自适应的观测模型,根据跟踪背景不同,采取最为合适的方式进行采样。并将这种改进的算法同传统的粒子滤波算法进行了实验对比。(3)针对当前对目标跟踪实时性的高要求,提出一种基于并行计算的粒子滤波跟踪算法,以当前计算机多核系统为保证,开发基于数据并行的分布式算法,将并行计算表示为一系列任务,任务之间通过使用通道发送消息进行相互通信。不仅能够提高跟踪精度,而且能有效地满足硬实时系统的时间约束问题。在集群环境中,设计并实现了粒子滤波跟踪的并行算法；采用组通信调度算法,最小化了每步通信所消耗的时间开销。对这种并行算法进行了仿真实验,同传统串行算法进行了数据对比,并针对这种算法进行海天背景的目标跟踪实验。更多还原

【Abstract】 Target tracking technology is one of the most popular research topics in the field of modern image processing and vision research. With the development of science and technology, especially the rapid development of computing power and communications technology, breakthrough improvement of the field of image recognition algorithm for target tracking technology to lay a solid theoretical foundation and the possibility of realization in recent years, target tracking technology astronomical prediction, the field of military defense, intelligent monitoring, traffic monitoring has important practical value.This article describes the research status of a ship in the sea and the sky background target tracking, technology and theory of traditional target tracking algorithm, the main work is the particle filter algorithm is applied to the sea and the sky target tracking, and tracking theory for the current applicable to the Haitian target tracking particle filter tracking algorithm, to make preliminary studies for the development of efficient real-time target tracking system for sea and the sky background, the specific content:(1) This paper summarizes several traditional target tracking theory Based on Kalman Filter and Mean-shift algorithm filter theory, experimental simulation of a video tracking, respectively, for these two algorithms, and Kalman Haitian target tracking filter algorithm error simulation, compare the Mean-shift algorithm, using Pap coefficient search target to match the initial frame target.(2) This paper summarizes the sea and the sky background, the traditional particle filter tracking algorithm principle, the singularity of feature selection for target tracking, the comprehensive utilization of the color histogram and texture histogram contains the target observation will after the fusion of color and texture of the target as a target tracking features, the establishment of an adaptive observation model, depending on the track background to take the most appropriate way of sampling. And this improved algorithm with the experiment compared to the traditional particle filter algorithm.(3) For the current high demand real-time target tracking, based on parallel computing particle filter tracking algorithm, the current computer multi-core systems in order to ensure the development of parallel distributed algorithm based on data parallel computation is expressed as a series of tasks, between tasks by using the channel to send messages to communicate with each other. Can not only improve the tracking accuracy, and can effectively meet the hard real-time constraints. In a cluster environment, the design and realization of the parallel algorithm of particle filter to track; group scheduling algorithm to minimize the time consumed by each step of communication overhead. A simulation of this parallel algorithms, data contrast with the traditional serial algorithm, target tracking experiments for this algorithm to the sea and the sky background.更多还原