摘要作为视频信息存储与追溯的重要手段，视频监控技术在军事及民用监控领域发挥越来越重要的作用。但由于视频信息量巨大，从大量视频中快速获取所需信息过程繁琐，尤其是在特定的类似无人值守的场合，往往需要对运动目标进行自动提取、识别、跟踪，并采取报警等后续措施，因此有必要开展基于视频信息流的目标自动识别技术研究。本课题主要针对相机运动状态下运动目标检测技术与方法开展研究工作，其主要难点可以归纳为以下三点：1）由视角变换引起的目标外观模型短时间内变化较大；2）在复杂的场景中，出现与目标特征相似的样本区域，导致错误跟踪；3）目标出现半遮挡或全遮挡的状态，导致跟踪丢失。针对以上问题，本文重点进行基于 Meanshift 的运动目标检测跟踪算法及实现，同时进行基于光流、基于粒子滤波的运动目标检测跟踪算法及实现，利用 Matlab软件对 3种算法进行仿真，达到检测与跟踪运动目标的效果，分析比较 3种算法的稳定性、准确性和实时性，提出今后研究的重点是针对 Meanshift 跟踪算法在目标外观模型变化时出现跟踪丢失、错误跟踪的问题进行算法改进。41327
毕业论文关键词 目标跟踪 均值漂移 光流 粒子滤波
Title Resasrch of Ground-target Stable TrackingAlgorithm in Video Surveillance System
Abstract As a important means of video information storage and traceability, videosurveillance technology has played an increasingly important role in the fieldof military and civilian control.However, due to the huge amount of videoinformation, it is tedious to get information quickly from a lot if vedio.Especially when in the certain occasions such as an unattended occasion, we oftenneed the video surveillance system to automatically extract, identify and trackthe moving target. Therefore, it is necessary to carry out the research of automatictarget recognition technology based on video stream. This topic mainly carry outresearch work for technology and methods of moving target detection under thecamera motion state. There are three main difficulties: 1)Target appearance modelchanges greatly in a short time resulted by perspective transformation; 2)Wrongtracking resulted by sample region which is similar to target feture in a complexscenario;3)Target is in the half or full occlusion state, which leads to loss oftracking. To solve the above problems, this topic focuses on moving targetdetection and tracking based on the principle of meanshift as well as that basedon the principle of optical flow and particle filter. In this topic, I use theMatlab software to simulate these three algorithms to achieve the effect of movingtarget detection and tracking, then analyse and compare their performance suchas accuracy，robustness and speed. This topic proposed that the main point of futureresearch is to solve the problem of wrong tracking in the Meanshift algorithm.
Keywords Target Tracking Meanshift Optical Flow Particle Filter
目 次
1 绪论  1
1.1 课题背景及研究意义  1
1.2 课题研究现状与发展趋势  1
1.3 运动目标检测跟踪技术研究中存在的问题  2
1.4 本文主要内容与结构安排  3
2 基于 Meanshift 的运动目标检测跟踪算法  5
2.1 基于 Meanshift 的运动目标检测跟踪算法原理及实现  5
2.2 基于Meanshift的运动目标检测跟踪算法的改进  10
3 基于光流的运动目标检测跟踪算法原理及实现  12
4 基于粒子滤波的运动目标检测跟踪算法原理及实现  15
4.1 目标模型描述  15
4.2 粒子样本集合描述  16
4.3 基于颜色直方图的粒子滤波算法  16
5 算法性能分析与比较  19
5.1 实验结果  19
5.2 性能分析  21
结论  22 Meanshift视频监控系统中地面目标稳定跟踪算法研究:http://www.751com.cn/tongxin/lunwen_41339.html