文档介绍:南京邮电大学
硕士学位论文
智能视频监控中的运动物体检测与跟踪算法研究
姓名:陈祥虎
申请学位级别:硕士
专业:计算机应用技术
指导教师:沈苏彬
2011-03
南京邮电大学硕士研究生学位论文摘要
摘要
监控环境的多样性和复杂性导致视频分析算法的鲁棒性较差,对于复杂的算法,在嵌
入式环境下也不能满足实时性需求。为了使得视频分析算法适合应用于嵌入式环境,本文
主要研究嵌入式环境下视频分析算法,算法的实现与优化,以及视频分析的应用接口。
嵌入式环境下的资源紧缺性决定了视频分析算法必须简单、高效和具有较好的鲁棒
性。本文研究了背景相减法、帧间差分法和光流法三种运动物体检测算法,通过理论分析
和仿真实验,选择背景相减法作为运动物体检测算法;在像素标记法的基础上通过改变图
像扫描方向,提出了一种改进的物体四边界定位算法,相比区域增长法和像素标记法具有
较低的时空复杂度;通过 Mean Shift 算法和 Kalman 滤波器相结合,有效地解决了快速运
动物体和物体大比例遮挡问题;在嵌入式环境下实现运动物体检测软件并且对代码进行优
化,使软件达到了实时性要求;通过研究 ONVIF 核心规范的视频分析应用结构、视频分
析配置接口和场景描述接口,设计和实现了嵌入式环境下的视频分析应用接口,对网络视
频处理器与视频分析处理器之间的视频流同步问题提出了一个技术方案,提供软件可配
置、网络视频处理器与视频分析处理器协同工作等能力。
关键词:运动物体检测,目标跟踪,视频分析,代码优化
I
南京邮电大学硕士研究生学位论文 ABSTRACT
ABSTRACT
Diversity plexity of surveillance environment leads to poor robustness for video
analysis algorithms. Complex algorithms can not meet real-time requirement under embedded
environment. In order to make video analysis algorithms suit for embedded environment, this
thesis mainly researches on video analysis algorithms, algorithms implementation and
optimization,and application interface of video analysis under embedded environment.
The scarcity of resources under embedded environment determines the video analysis
algorithms must be simple, efficient and robust. This thesis makes a research on background
subtraction, frame difference and optical flow for moving object detection algorithm. Through
theoretical analysis and experimental simulation, background subtraction is selected as moving
object detection algorithm. An improved locating algorithm for the four boundaries of a object
is proposed on the basis of pixel-labeled algorithm by changing the direction of image
pared to region growing algorithm and pixel-labeled algorithm, the improved
algorithm has a lower temporal and plexity. Mean Shift bining with
Kalman filter provides an effective solution to tracking of fast moving objects and objects with
large proportion