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:
地物背景下运动目标检测与跟踪技术研究
作者: 赖作镁
学位类别: 博士
答辩日期: 2007-05-31
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: 王敬儒
关键词: 目标检测与跟踪 ; 电子稳像 ; 背景运动补偿 ; Kalman滤波 ; 粒子滤波
其他题名: Research on the Technique of Moving Target Detecting and Tracking in Complex Environment
学位专业: 信号与信息处理
中文摘要: 地物背景下运动目标检测与跟踪技术是光电成像跟踪系统的关键技术之一,论文围绕地物背景下运动目标检测和稳定跟踪技术的理论、算法以及工程实现展开深入细致的研究,并提出了解决方案。 针对地物背景非常复杂、目标机动性强的特点,传统的基于单帧的静止图像分割算法一般无法检测目标,因此,论文首先分析了导致图像帧间灰度变化的影响因素,在成像系统搜索目标阶段,如果不考虑光照变化影响,可以认为摄像机运动是地物背景下帧间图像灰度变化的主要原因。在深入分析摄像机成像的物理模型的基础上,通过分析三维空域运动场到二维图像速度场的投影转换模型,得出目标检测中可以用仿射变换模拟场景运动的结论,为后续的算法研究提供了理论依据。 为了改善光电成像跟踪系统中视频的视觉质量,以便后续的目标跟踪,论文在详细对比传统电子稳像方法的基础上,提出了基于Kalman滤波的电子稳像算法。该方法利用特征点对应法求相邻帧间的运动,然后用最小二乘法计算出全局的仿射运动参数,最后,Kalman滤波方法把运动随时间的变化视为随机过程,把抖动参数当成运动的噪声,通过对抖动参数的分析从全局运动参数中分离出摄像机的正常扫描运动参数,达到稳像的目的,并用峰值信噪比和均方误差两个指标定量分析了算法的有效性。 充分利用背景与目标的运动特性,提出了基于背景运动补偿的目标检测算法。该算法采用多分辨率参数估计分层精确地估计摄像机的仿射运动参数,在各个分辨率下,运用Levenberg-Marquardt迭代法进行参数调整。背景运动补偿算法使摄像机运动情况下的目标检测问题转化为摄像机静止下的目标检测问题。针对差分检测法中的阈值选择问题,提出假设检验方法成功检测出地物背景下的运动目标。 地物背景下运动目标的稳定跟踪是本论文研究的又一个重点。针对经典模板匹配跟踪算法的局限性,论文对粒子滤波理论进行了深入的分析,在此基础上提出了新颖的粒子滤波目标跟踪框架;该框架把目标跟踪问题看成状态估计问题,采用基于贝叶斯推理和蒙特卡罗方法的粒子滤波算法递归地实现了地物背景下的目标跟踪。在此框架的基础上,首先采用一种非参数估计的方法,引入核函数直方图分布作为目标观测,融入粒子滤波跟踪算法当中。模拟运动目标跟踪实验证明了粒子滤波跟踪方法的鲁棒性和稳健性,相比于模板匹配方法,具有明显的优势。真实场景跟踪实验证明了基于核密度直方图的目标观测具有良好的抗光照变化能力。 针对地面目标特征发生严重变化情况下仍能保持稳定跟踪的跟踪算法的两个主要要素——目标观测模型的描述和目标模板更新策略进行了细致分析,并提出了有效的融合解决方案,即把混合高斯模型引入粒子滤波的目标跟踪框架中,目标的观测模型采用含有三个分量的混合高斯模型来建模并用EM算法在线自适应地更新,形成基于混合高斯模型的粒子滤波算法,从而很好地解决了模板更新问题,该算法在地物背景下目标快速旋转、目标外观姿态快速变化和局部遮挡情况下的跟踪均取得理想的效果,且算法不需要普通粒子滤波算法的重采样策略,计算简单,易于硬件并行实现。 在基于CPCI架构的FPGA+DSP并行实时处理平台上,对基于混合高斯模型的粒子滤波实时跟踪算法进行优化,用FPGA实现了该算法,实验结果表明该算法能满足复杂背景下运动目标实时跟踪要求。 总之,论文针对光电成像跟踪系统中地物背景下的目标检测与稳定跟踪问题,结合工程实际,提出了一些针对性很强的思路和解决方案,为提高光电成像跟踪系统的水平奠定了基础。
英文摘要: Moving target detection and tracking in complex environment have been the key technique of the optical-electro detecting system. Aiming at the key issues of the dissertation, including theory, algorithm and hardware implementation of moving target detection and steady tracking, elaborately analysis and study were performed. The main objective of this dissertation is to develop some effective algorithms to solve some difficulties. Based on the study of the characteristics of target that image background is always complex and changeful while target’s motion is highly maneuverable, target is difficulty to be detected with the general still image segmentation methods. Our research started with dynamic image analysis theory, at the stage of target searching, without regard to the change of illumination condition, the reason that resulting in change of image intensity between frames was exhaustively discussed, and educed an important conclusion that camera motion and dithering were the main reason of image intensity change. The transformation model of 3D motion field to 2D image velocity field on the basis of camera model is seriously analyzed, and obtained an useful conclusion that affine model can well simulate scene motion in the target detection system, which provided theory for later algorithms. In order to improve video quality and performance of target tracking, after a series of experimentation on traditional methods, a new electronic stabilization algorithm was proposed based on recursive Kalman filter and global motion estimation. Motion vector was detected by feature point extraction and corresponding algorithm. Affine motion parameters were calculated through least square method. Recursive Kalman filter was applied to separate the dithering parameter from global motion parameter through considering camera motion as a stochastic process. Experimental results indicate the algorithm is effective and robust after extensive experiments over a wide variety of videos. In the other hand, PSNR and MSE showed quantificationally that the proposed algorithm is effective. Moreover, real time process requirement was satisfied by the algorithm. As for target detection, a new algorithm based on background motion compensation was put forward that affine model simulates camera motion in the scene, and then obtain a precise motion compensation frame after robust estimation affine parameters through multi-scale motion estimation algorithm, where affine parameters were adjusted by Levenberg-Marquardt algorithm at each resolution, which translated target detection problem of moving camera into the problem of still camera through extruding target and removing background. Difference image after compensation was detection through hypothesizing and testing. After a series of morphological process, we obtained intact moving object. Experimental results indicated that this algorithm can effectively eliminate background motion and segment object in complex environment. Robust and steady target tracking is another important issue of this dissertation. After analyzing the shortcoming of general template match tracking, a new tracking framework was constructed based on particle filter, which regarded nonlinear and non-Gaussian target tracking as state estimation parameter, particle filter implemented the process though Bayesian reasoning and Monte Carlo theory. According to the framework, kernel histogram was introduced as a measure model and integrated into particle filter tracking algorithm. Compared with template match tracking through simulated moving target track experiments, the new algorithm had a good advantage over template match algorithm through bartering a litter bigger computation and approximate track error for steady target tracking, moreover, it can also track the target with variational illumination owing to kernel histogram measurement. As for two main problems in robust and steady target tracking, the description of target measure model and the technique of template update were meticulously studied. Mixture Gaussian model was introduced into particle filter tracking framework to be an effective target description, which was compose of three components and online updated with incremental EM algorithm. Many experiment results show that the algorithm can also track targets under variational illumination, especially for fast varying appearance and pose and local occlusion, furthermore, it is simple enough to be a real-time object tracking algorithm without resample. At last, target detection and tracking platform based on DSP+FPGA with CPCI architecture was introduced. The particle filter tracking algorithm was implemented on FPGA through algorithm optimization. Experiment data proved the real-time algorithm. In conclusion, some innovated methods were developed to solve the difficulties for target detection and steady tracking in complex environment. The work of this dissertation has important reference to the future development of the optical-electro detecting system.
语种: 中文
内容类型: 学位论文
URI标识: http://ir.ioe.ac.cn/handle/181551/227
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_200418015127021赖作镁_paper.pdf(7096KB)----限制开放View 联系获取全文

Recommended Citation:
赖作镁. 地物背景下运动目标检测与跟踪技术研究[D]. 光电技术研究所. 中国科学院光电技术研究所. 2007.
Service
Recommend this item
Sava as my favorate item
Show this item's statistics
Export Endnote File
Google Scholar
Similar articles in Google Scholar
[赖作镁]'s Articles
CSDL cross search
Similar articles in CSDL Cross Search
[赖作镁]‘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_200418015127021赖作镁_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