基于计算机视觉的目标检测和阴影检测算法的研究

发布时间：2018-05-01 21:17

本文选题：阴影检测 + 混合高斯模型　；参考：《吉林大学》2017年硕士论文

【摘要】：计算机视觉是一种研究如何使计算机能够从数字图像或者视频中获得高级理解的交叉学科领域,从工程的角度来看,它试图自动化人眼视觉所能做的任务。计算机视觉的工作包括获取、处理、分析和理解数字图像的方法,以及从真实世界进行高维数据的提取来产生诸如决策形式中的数字或符号信息。作为一个学科,计算机视觉与从图像中提取信息的人造系统背后的理论有关;这些图像数据能够具有多种形式,诸如视频序列,多个摄像机视角,或者从一个医学扫描仪获得的多维数据。作为一个技术学科,计算机视觉试图应用它的理论和模型来进行计算机视觉系统的构造。智能视频监控是利用计算机视觉技术对视频信号进行处理、分析和理解,在不需要人为干预的情况下,通过对序列图像自动分析监控和定位场景中变化的技术。智能视频监控系统可以从繁复的视频帧中分辨、识别感兴趣的目标物体,自动分析、提取视频源中的丰富的有用信息,进而改善了传统视频监控系统的智能化水平。随着网络技术和信息技术的高速发展,模式识别领域中智能视频监控技术得到了越来越多的关注程度,已经逐步成为近年来的最热门的科学研究主题之一。在智能视频监控领域中,运动目标的检测是至关重要的研究课题,传统的运动目标检测技术包括光流法,帧差法和背景差分法,这些算法能够有效的检测出运动的目标。但是当前的前景检测技术通常具有一个重大的缺陷:阴影往往被分类成为前景的一部分。这种情形的发生是因为阴影与产生阴影的运动目标共享相同的运动模式并且具有相似的亮度变化,这些阴影被误的分类为前景导致了不精确的检测结果并降低了跟踪的性能,因而,如何获得干净准确的目标便成为了必不可少的一个环节。本文的研究内容主要包括对混合高斯模型的改进并且提出了一种融合纹理特征和颜色特征的阴影检测算法。由于混合高斯模型算法收敛速度缓慢,而且采用统一的更新速率使得背景模型在某些情况下耗时且背景模型不能及时的反映真实的背景变化,本文提出了一种改进的混合高斯模型,并在背景更新过程中在不同的时间段使用不同的学习率,提高了背景建模收敛的速率,也提高了感兴趣目标的检测率。同时,本文提出了一种融合颜色和纹理特征的阴影检测算法,在投射阴影检测的过程中,基于颜色特征的阴影检测得到了广泛的应用,但当运动目标与背景相似时运动目标常会被误检为阴影,为了解决这个问题,结合纹理特征,本文提出了一种基于融合颜色和纹理的阴影检测算法。仿真实验也表明算法的有效性。
[Abstract]:Computer vision is an interdisciplinary field that studies how to make computers gain advanced understanding from digital images or video. From the point of view of engineering, it attempts to automate the tasks that human vision can do. The work of computer vision includes the acquisition, processing, analysis and understanding of digital images, as well as the extraction of high-dimensional data from the real world to produce digital or symbolic information such as decision forms. As a discipline, computer vision has something to do with the theory behind artificial systems that extract information from images; these image data can take many forms, such as video sequences, multiple camera views, Or multidimensional data from a medical scanner. As a technical subject, computer vision tries to construct computer vision system by applying its theory and model. Intelligent video surveillance is a technology that uses computer vision technology to process, analyze and understand the video signal. It can automatically analyze, monitor and locate the changes in the scene by analyzing the sequence images without human intervention. The intelligent video surveillance system can distinguish the object of interest from the complicated video frame, automatically analyze and extract the rich useful information from the video source, and then improve the intelligence level of the traditional video surveillance system. With the rapid development of network technology and information technology, intelligent video surveillance technology in the field of pattern recognition has been paid more and more attention, and has gradually become one of the hottest scientific research topics in recent years. In the field of intelligent video surveillance, moving target detection is an important research topic. Traditional moving target detection techniques include optical flow method, frame difference method and background differential method. These algorithms can effectively detect moving targets. However, the current foreground detection techniques usually have a major defect: shadows are often classified as a part of the foreground. This happens because shadows share the same motion pattern and have similar luminance changes as the moving objects that produce them. These shadows are wrongly classified as foreground resulting in inaccurate detection results and reduced tracking performance. Therefore, how to achieve a clean and accurate goal has become an essential link. The main contents of this paper include the improvement of mixed Gao Si model and a shadow detection algorithm combining texture features and color features. Due to the slow convergence rate of mixed Gao Si algorithm and the uniform updating rate, the background model is time-consuming in some cases and the background model can not reflect the real background changes in time. In this paper, an improved mixed Gao Si model is proposed, and different learning rates are used in the process of background updating, which improves the convergence rate of background modeling and the detection rate of objects of interest. At the same time, a shadow detection algorithm combining color and texture features is proposed in this paper. In the process of shadow projection detection, shadow detection based on color features has been widely used. However, when the moving object is similar to the background, the moving object is often mistakenly detected as a shadow. In order to solve this problem and combine the texture features, a shadow detection algorithm based on the fusion of color and texture is proposed in this paper. Simulation results also show the effectiveness of the algorithm.
【学位授予单位】：吉林大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.41

【相似文献】