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:
嵌入式实时多目标识别跟踪系统设计及识别算法研究
作者: 冷何英
学位类别: 博士
答辩日期: 2003
授予单位: 中国科学院光电技术研究所
授予地点: 中国科学院光电技术研究所
导师: 王敬儒
关键词: 多目标识别 ; 多特征融合 ; 高速DSP ; 嵌入式实时系统 ; 并行处理
其他题名: Research of the Embedded Real-Time Multiple Target Recognition Tracking System Designing and Recognition Algorithms
中文摘要: 现代战争中大量空袭武器多方向、多批次地攻击,使得多目标识别跟踪技术成为现代OE设备发展的关键技术之一。电视多目标识别跟踪技术是一门新型的探测技术,由于起步较晚,与成熟的红外、雷达多目标技术相比存在着许多技术难题。因此,本论文将针对实际电视捕获跟瞄系统工程实现中出现的技术难点,又结合国家863-802高技术预研课题的要求,从嵌入式实时处理平台开发和目标识别方法两方面进行了深入的研究与实践。本文首先通过分析我所现有基于ADSP2183的视频处理系统的局限性和新型嵌入式实时多目标识别跟踪系统的基本功能,确定了“以TI最新推出的超高速DSP芯片C6202为核心处理器、采用DSP+FPGA+CPLD的紧祸合可重构互连体系结构”的系统总体方案,重点进行了软、硬件平台的开发和系统集成实验,深入研究了高速嵌入式DSP系统实现的软硬件优化策略。该新型实时平台的成功开发使得我所原有的视频处理平台在处理速度,体系结构、任务分配、功能实现等方面都有了显著的提高和改进,实现了平台的全面升级。本文在单高速DSP嵌入式系统成功开发的基础土沟苗准视频跟踪系统高顷频采集、多通道传输、多传感器融合和夏杂算法工程化的发展方向,结合国内外DSP并行处理系统的发展现状,确定了“适用迁高J颐频多通道CCD的CompactPCI架构并行处理系统”的总体方案和体系结构,重点进行了各关键子板的硬件设计和并行子任务的划分,提出“双专用总线”、“两级预处理”、“功能模块化”等一系列设计新思想,为我所今后实现高速并行处理系统工程化提供了技术储务。在多目标识别方法研究方面,一首先通过对电视识别跟踪系统中各类成像目标的特性比较,深入分析了远近探测距离下的多目标特性,并针对点状多目标的宏观整体特性,建立了点源日标模型、噪声统计模型、图像序列模型;针对宏状多目标的微观精细特性,建立了成像模型、仿射变换模型、多特征复合目标链识别模型,对后续识别准则的建立及识别算法的设计提供了准确的数理模型描述。本文根据各目标之间的距离远近,分别研究了点状多目标在分离和粘连两种情况下的识别方法。针对前者,提出了一种基于距离相似度和嫡相似度的粗精两级双搜索识别算法,利用基于距离相似度的目标粗分割、快速多目标编目、基于最大局部熵的目标精识别和改进区域生长法,实现了分离多目标的精确定位和目标边缘重构;针对后者,提出一种基于极限腐蚀思想的静态识别算法,有效地避免了动态多目标跟踪方法存储量过大、轨迹预测困难的局限性。此外,本文还从人类视觉系统收集和处理信息的选择性和层次性出发,提出基于知识基的由粗到精的宏状多目标分层识别策略。预处理中,通过基于隶属度函数变换的模糊边缘检测算法,减少了强杂波干扰和非均匀性照度带来的低灰度图像边缘信息的损失,实现了图像增强;在多特征提取层中,首次定义了广义边界链码的概念,并提出一种新的基于USAN原则和广义边界链码的粗精两级角点检测算法。实验验证该算法融合了目标的区域特征和边缘特征,具有快速、高效、局部抗噪能力强、检测精度高等特点。然后在模糊边缘和角点检测的基础上,抽取了宏状目标的2个欧式不变量一角点夹角、矢量边长和1个仿射不变量一边缘矩,并首次将欧式几何中的相似性角定理应用于形态特征明显的多目标分类识别中;在目标分类层中,引入在线辨识的新概念,设计了多特征链复合和逐级判断的MCF目标分类器,从而既保证了目标分类的正确率,又加快了分类执行速度。
英文摘要: With many weapons of modern war needing to attack the given targets in many directions or in many orders, multi-target recognition and tracking becomes a very important technology in the development of model optical-electronic equipments. Compared with infrared multi-target and radar multi-target, TV multi-target recognition and tracking is a new detection technology, and many problems will be met in its performance process. So the paper researches two problems in designing a new embedded real-time processing platform and presenting some target recognition approaches. Firstly, according to analyzing the limit of the old real-time video processing system based on ADSP2183 of our institute, the basic function and structure of a new embedded multi-target recognition and tracking system is proposed. It is based on a high-speed core processor TMS320C6202 and DSP+FPGA+CPLD structure. After system software and hardware platform being designed in detail, real-time test experiments of the system are finished successfully and many optimization approaches on software programing and hardware implementation are discussed. Being compared with old system, our new system is improved evidently on processing speed, reasonable task being assigned and function implementing. Secondly, because modern video tracking system must be satisfied with sampling image at high frame rate, transmitting data through multiple channels and fusing all kinds of information from multiple sensors, a new parallel processing system based on CompactPCI specifications for a high-frame rate and multiple channels CCD is designed and proposed. Then all important boards are designed and parallel processing tasks are programmed, some new ideas such as two special buses, two pre-processing function modularization are presented in the system design. It provides certain technologies for project implementation of high-speed DSP parallel system in the future. Thirdly, according to being comparing with all kinds of imaging target features in the TV recognition and tracking process, multi-target features of the different detection distance are analyzed in detail. For the point multi-target, three important mathematical physical models based on its macro and unitary features such as point target model, noise statistics model and image sequence model are set up. And for the macro shape multi-target, three important mathematical physical models based on its microcosmic and detail feature such as imaging model, affine transform model and multi-feature fusion recognition model are also built. The exact description of these models is foundation of designing recognition rules and proposing recognition algorithm. Because the space of each point targets in the far distance is different, the paper studies the recognition methods of point targets on the separate or conglutinative condition. For separate point targets, a new algorithm based on distance conformability degree principle and entropy conformability degree principle is proposed to detect targets from rough to precision. It utilizes many ways such as rough target segmenting, fast multi-target listing, precise target location based on maximal local entropy value of image and improved region growing to get the precise positions of multiple targets and reconstruct target boundary. For conglutinative point targets, a static recognition algorithm based on ultimate erode idea is utilizes to overcome effectively such shortages of dynamic multiple target tracking algorithm as trace forecast being difficult and data storage being too much. Finally, according to the selectivity and hiberarchy of information being collecting and processing in the human vision, a multi-layers target recognition strategy based on pre-knowledge is proposed. In the pre-processing layer, an edge extraction algorithm through membership function transform is utilized to reduce information losing of low-contrast image edge and enhance image signal being disturbed by strong background clutter and non-uniform illumination. In the multi-feature extracting layer, a general conception of boundary chain code is developed for the first time. Then a new multi-stage corner detection algorithm based on USAN decision rule and general boundary chain code is proposed. The experiment results indicate the algorithm utilizes region feature and edge feature of targets to increase corner detection speed and precision, reduce local noise effect. After corners and edge detection of macro shape targets are extracted, corner angle and vector length based on geometry invariants and boundary moments based on affine invariants are computed. And the similarity angle theory of Euclidean theories is applied to macro shape multi-target classification for the first time. In the target classification layer, a new concept named on-line recognition is developed and a new target classifier named MCF is constructed based on multiple feature fusion and judging by degrees. The emulation results show the recognition rate and classification speed of macro shape targets are improved greatly.
语种: 中文
内容类型: 学位论文
URI标识: http://ir.ioe.ac.cn/handle/181551/86
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
LW011655.pdf(2185KB)----限制开放View 联系获取全文

Recommended Citation:
冷何英. 嵌入式实时多目标识别跟踪系统设计及识别算法研究[D]. 中国科学院光电技术研究所. 中国科学院光电技术研究所. 2003.
Service
Recommend this item
Sava as my favorate item
Show this item's statistics
Export Endnote File
Google Scholar
Similar articles in Google Scholar
[冷何英]'s Articles
CSDL cross search
Similar articles in CSDL Cross Search
[冷何英]‘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: LW011655.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