基于背景差分法的行人检测研究 - 图文

2016届计算机科学与技术专业毕业设计（论文）

毕业设计（论文）

题 目 基于背景差分法的

行人检测研究

专 业 计算机科学与技术 班 级 学 生 指导教师

2016 年

刘姗姗：基于背景差分法的行人检测研究

摘 要

现阶段许许多多的监控类系统的运作进行方式依然需要很多相关工作人员不断地监视和分析视频内的场景，不仅没有休息时间，而且工作量特别大，实时性信息和监控实时性两者矛盾越来越明显。这便是研究本课题的意义，本论文阐述了本课题的研究现状、实现它所涉及到的背景建模方法、特征提取方法和背景更新等。

此课题研究的内容是基于背景差分法行人的视频监控系统：在不需要人为进行干预的前提下，利用视频分析和机器学习等的一些方法，通过针对于截取到的图像进行实时的场景以及目标物做出有可行性的分析，将摄像机截取到的视频图像的帧通过自动化的检测分析处理，进而达到对动态的场景中目标的行以及其和位置进行确定的目的，然后进行跟踪检测，并且对于检测到的目标的行为进行分析、语义的描述以及理解，实现模仿与人类相似的生物视觉和分析的功能。但由于很多的方法还是不够成熟和稳健，因此有必要做大量的研究以提高在较为复杂的环境中检测运动目标的成熟度和准确率，进而提高实用化的效果。

此程序是以MATLAB进行实现，开始利用了混合高斯模型从当前复杂的背景图像序列中构建背景模型，此时首先需要使用方差、均值以及权值来进行计算建模所需要的数据，如马兹距离等，在此过程中，方差尽量大一些，权值尽量小一些，这样有利于获得高斯模型；然后据此模型采用联通区域轮廓特征进行提取

2016届计算机科学与技术专业毕业设计（论文）

行人特征。在以上过程中以加权更新的方法构建实时的模型更新机制，从而规避了背景被污染。

关键字：混合高斯，联通区域，特征提取，加权更新

刘姗姗：基于背景差分法的行人检测研究

ABSTRACT

To present many monitoring of such systems works still need a lot of related staff continually monitoring and analysis in the video scene, not only no time to rest, and particularly large workload, real-time information and real-time monitoring is the contradiction is more and more obvious. This is we attentively studies the significance of this topic, this paper expounds this topic research present situation, implement it involved background modeling method, the feature extraction method and background, etc.

My study is the content of the video monitoring system based on background difference method pedestrians: on the premise of without the need for human intervention, using some methods of video analysis and machine learning and so on, through to the image on the real-time scene make a feasibility analysis, and target of the camera to capture the video image frame through the analysis of the test automation processing, thus achieve the goal of the dynamic scene line as well as its purpose, and location to determine and track detection, and to analysis the behavior of the detected target, the semantic description and understanding, imitation is similar to human biological vision and the analysis of the function. But as a result of a lot of methods are not enough mature and steady, so it is necessary for us to do a lot of research in order to improve the moving target detection in complex environment of maturity and accuracy, and improve the effect of practical application.

2016届计算机科学与技术专业毕业设计（论文）

This program based on matlab implementation, began using the gaussian mixture model from the current complex background image sequences to construct background model, first you need to use at this time, the mean and variance weights to calculate needed for modeling data, such as horse, distance etc., in the process, the variance as far as possible a few bigger, weights of less as far as possible, that is conducive to gain gaussian model; Then according to this model USES the unicom area contour features extracted to pedestrians. In the process of the above with weighted update method to build model of real-time update mechanism, so as to avoid the contamination in the background. Key

words:mixed

gaussian

，

connected

regions,feature

extraction,weighted update

本文来源：https://www.bwwdw.com/article/y048.html