中国科学院光电技术研究所机构知识库
Advanced  
IOE OpenIR  > 光电技术研究所博硕士论文  > 学位论文
题名:
基于多特征的目标识别与跟踪技术应用研究
作者: 宋建勋
学位类别: 博士
答辩日期: 2009-06-04
授予单位: 中国科学院光电技术研究所
授予地点: 光电技术研究所
导师: 吴钦章
关键词: 目标识别 ; 目标跟踪 ; 信息融合 ; 模糊综合函数 ; D-S证据理论 ; 并行处理
其他题名: Application Research of Target Recognition and Target Tracking Technology Based on Multiple Features
学位专业: 信号与信息处理
中文摘要: 光电跟踪测量系统在天文、空中交通管理、无人机、安全监控系统、自动驾驶系统以及智能机器人等方面都有着广泛的应用。随着科学技术的发展,各个领域对光电跟测量踪系统提出了更高的要求,要求在尽可能远的距离上能够对目标进行识别与跟踪,并具有较高的跟踪精度和跟踪可靠性。本论文根据光电跟踪测量系统工程应用的要求,将光电跟踪测量系统的跟踪关联模块进行了嵌入式实时处理平台开发,并对光电跟踪测量系统的目标识别与跟踪算法进行研究。 跟踪关联模块是光电跟踪测量系统的一个重要组成部分,它主要对探测到的目标进行关联处理,得到各个目标的航迹。跟踪关联模块是一个复杂的数据处理过程,数据处理量比较大,因此确定了以TI最新推出的高性能浮点DSP芯片TMS320C6713为核心处理器,采用主机+PCI+FPGA+双DSP的紧耦合可重构互联体系结构的总体方案,将跟踪关联模块在双DSP芯片上运行,形成一个独立的跟踪关联嵌入式系统。论文从跟踪关联嵌入式系统的硬件和软件平台进行了详细的分析,对跟踪关联嵌入式系统平台中的DSP、SDRAM、FLASH等器件进行了调试,重点对跟踪关联系统在DSP上的代码进行了优化,将跟踪关联模块在DSP中的运行时间从最初的110多毫秒减少到20毫秒以内。最后对并行处理系统的三个要素:处理单元的选择、并行系统互联结构、并行算法程序和任务分配原则进行了深入的分析与研究,并给出了几种典型的多DSP并行处理系统结构,为今后光电跟踪测量系统嵌入式多DSP高速并行处理平台的设计和复杂算法的工程实现奠定了基础。 目标跟踪的算法有很多,许多算法发展的也比较成熟,但是一般都比较复杂,工程中实现起来比较困难,无法满足实时性的需要。针对图像目标跟踪的特点,从图像中提取目标的多种特征信息,充分利用这些信息来提高目标跟踪的精度和跟踪的可靠性。论文分析了在工程中应用的最近邻数据关联算法,实现了最近邻数据关联算法中预测模块的最小二乘递归预测,深入研究了位置关联、多特征关联的实现过程,通过多特征关联可以在位置关联的基础上进一步剔除虚假目标,提高目标跟踪的精度。最后对暂时航迹按照置信度进行排序,删除无效的航迹,减少航迹存储的开销,同时提高了航迹关联的效率。 在目标识别方面,主要应用了基于信息融合的目标识别方法。从光电跟踪测量系统的图像中提取多种特征信息,一方面这些特征信息能够从不同的角度更准确地反映目标的本质特征,通过信息融合的方法,把多种特征信息综合为一种联合特征,从而更有效地进行目标识别;另一方面这些特征信息往往具有不精确性、不确定性,甚至是模糊的、彼此不一致的和时变的,而模糊理论是解决这类问题强有力的数学工具,在模式识别方面应用比较广泛。因此论文提出了基于模糊综合函数的多特征融合识别方法,首先构造了特征的模糊化隶属函数,再对目标特征进行模糊化处理,然后根据模糊相似性公式计算出各个目标样点的模糊化判决结果,最后采用模糊综合函数对各个目标的模糊化判决结果进行综合处理得到各个目标点的可能性分布,再根据一定的模糊模式识别规则给出目标的判决结果。该方法不仅实现了不同特征之间的融合,而且也实现了特征在时间域上的融合,充分利用了各种信息,实验结果表明该算法能够有效地对光电跟踪测量系统图像中的目标进行识别。此外,论文还研究了基于D-S证据理论的多特征融合方法,D-S证据理论具有较强的理论基础,它可以明确的区分和处理信息的不确定性和不精确性。针对基本概率分配函数比较难以获取的难题,通过分析模糊理论的隶属函数和D-S证据理论的基本概率分配函数的关系,提出了利用高斯隶属函数构造基本概率分配函数的方法,然后再根据合成规则将多种特征进行合成,最后根据D-S证据理论的决策规则识别出目标。这种方法有机地将D-S证据理论、模糊理论和信息融合技术相结合,并用于目标识别。实验结果表明通过信息融合的方法能够有效地进行目标识别,将多种理论和信息融合方法相结合是当前目标识别研究领域的一个热点。
英文摘要: The electro-optic tracking and measuring system has a wide application in the field of astronomical observation, air traffic managerment, unmanned aerial vechicle(UAV), monitor system, auto-navigation system and intelligent robot. With the development of science technology, the electro-optic tracking and measuring system must recognize target and track target in a long distance, and it is necessary to meet the requirement of high accuracy and reliability. According to the requirement of the project, the dissertation for PhD degree presents the process of development of the embedded real-time processing platform, and researches on the algorithms of the target recognition and target tracking of the electro-optic tracking and measuring system. The tracking association module is an important component of the electro-optic tracking and measuring system and it implements the function of the data association and acquires the trajectory of every target. Because the tracking association module is a complicate data processing algorithm in nature, it is based on the high performance float-point core processor TMS320C6713 and host+PCI+FPGA+two-DSP structure, which is the tightly coupled and reconfigurable embedded platform. After debugging of the embedded system including DSP, SDRAM, FLASH, it is emphasized to optimize the code,then the tracking association module is run in the embedded platform. The run time is declined from about 110 milliseconds to within 20 milliseconds greatly. The dissertation also presents the software and hardware of the embedded system in detail. There are three fundamental elements that are close and independent in multi-DSP parallel system: processing unit, parallel structure of system, parallel algorithm and its task distribution, and they are described in detail, The dissertation researches several typical multi-DSP parallel processing structures, and it lays a foundation for the design of multi-DSP parallel processing platform and the project application of the complex algorithm in the electro-optic tracking and measuring system in the future. There are many target tracking algorithms which are efficient but too complex commonly, thus it is difficult to implement in the project. According to the image target characteristic, many features can be acquired from the image. Then, they are made full use to improve the tracking accuracy and the tracking reliability. The dissertation analyzes the nearest-neighbor data association algorithm which is used in the project, and implements the least square forecast algorithms by the recursion mode, and then researches deep the process of the position association and the multi-feature association. The multi-feature association can delete the false targets on the basis of the position association to improve the tracking accuracy dramatically. All trajectories are ordered in terms of the respective belief degree, and the noneffective trajectories are deleted, so the memory space is reduced and the efficiency of the trajectory association is improved. The target recognition methods based on the information fusion are presented in this dissertation. Multiple features of the image from the electro-optic tracking and measuring system can be acquired. On one hand, multiple features can express the target from different aspects, so it promises to integrate these features into a synthetical feature for the target recognition by the information fusion technique; On the other hand, the feature from the electro-optic tracking and measuring system are imprecise, uncertainty, fuzzy, inconsistent and changed. Fuzzy theory is a powerful mathematics tool to resolve the problem; therefore it has a wide application in the pattern recognition. The dissertation presents the algorithm of multi-feature information fusion based on fuzzy integration function. At first, the feature from sequence image is dealed with by fuzzy remembership function, and then fuzzy results are calculated by fuzzy similarity formula. At last, fuzzy results are processed compositively by fuzzy function, and the probability distribution is acquired, and then the result of target recognition is got by the fuzzy recognition rule. The algorithm makes full uses of all kinds of feature information, and it implements the fusion not only among different features but also in time.Subsequently, the dissertation presents the algorithm of multi-feature information fusion based on D-S theory of evidence. The algorithm has strong theory foundation, and it can differentiate and process the uncertainty and imprecision. It is difficult to acquire the basic probability assignment function (BPAF), and the fuzzy membership function and BPAF has some relationship in some degree by analysis, so the dissertation presents the method that gauss membership function constructs the BPAF, and then fuses features according to the composition rule, and recognize the target by the decision rule. The method combines the D-S evidence theory and the fuzzy theory with the information fusion technique for the target recognition. It is shown that the method using the information fusion technique can recognize target effectively, and it is a hotspot in the field of the target recognition to combine many theories with the information fusion technique.
语种: 中文
内容类型: 学位论文
URI标识: http://ir.ioe.ac.cn/handle/181551/354
Appears in Collections:光电技术研究所博硕士论文_学位论文

Files in This Item:
File Name/ File Size Content Type Version Access License
10001_200618015127035宋建勋_paper.pdf(1559KB)----限制开放View 联系获取全文

Recommended Citation:
宋建勋. 基于多特征的目标识别与跟踪技术应用研究[D]. 光电技术研究所. 中国科学院光电技术研究所. 2009.
Service
Recommend this item
Sava as my favorate item
Show this item's statistics
Export Endnote File
Google Scholar
Similar articles in Google Scholar
[宋建勋]'s Articles
CSDL cross search
Similar articles in CSDL Cross Search
[宋建勋]‘s Articles
Related Copyright Policies
Null
Social Bookmarking
Add to CiteULike Add to Connotea Add to Del.icio.us Add to Digg Add to Reddit
文件名: 10001_200618015127035宋建勋_paper.pdf
格式: Adobe PDF
此文件暂不支持浏览
所有评论 (0)
暂无评论
 
评注功能仅针对注册用户开放,请您登录
您对该条目有什么异议,请填写以下表单,管理员会尽快联系您。
内 容:
Email:  *
单位:
验证码:   刷新
您在IR的使用过程中有什么好的想法或者建议可以反馈给我们。
标 题:
 *
内 容:
Email:  *
验证码:   刷新

Items in IR are protected by copyright, with all rights reserved, unless otherwise indicated.

 

 

Valid XHTML 1.0!
Copyright © 2007-2016  中国科学院光电技术研究所 - Feedback
Powered by CSpace