基于视频的室内人体简单行为分析-通信与信息系统专业论文.docxVIP

摘 要 基于视频的人体行为分析是计算机视觉的研究热点。本文主要对目标检测、 人体跟踪和行为描述与理解这三个模块进行了研究，并开发了一个基于视频的人 体行为分析系统。该系统能自动检测、跟踪视频范围内出现的人体（一个或两个）， 并对每个人体的简单行为进行识别，且能统计出每个人体坐在座位的时间。本文 主要工作如下： 1.目标检测。本文研究了几种常用的目标检测算法和较新的 VIBE 算法，并 将它们进行比较，实验证明 VIBE 算法相比其它常用算法取得较好效果。但是该 算法不能适应于准静止目标的检测，针对该问题本文结合单高斯模型对 VIBE 算 法进行了改进。最后，将本文提出的算法和几种常用目标检测算法以及 VIBE 算 法进行了实验比较与分析。结果表明，本文提出的算法比几种常用目标检测算法 检测效果更好，与 VIBE 算法相当，并且能检测出准静止的目标。 2.人体跟踪。遮挡处理问题一直是人体跟踪中的热点和难点问题，现有的跟 踪算法大多数只能处理人体的部分遮挡或者短暂遮挡，对于完全遮挡和持久遮挡 效果并不理想。本文采用了基于检测匹配的人体跟踪算法，并结合一种遮挡处理 算法来处理人体的持久遮挡问题。实验证明，通过把基于检测匹配的人体跟踪算 法和该遮挡处理算法相结合，取得了比较好的跟踪效果且能处理人体的持久遮挡。 3.行为描述与理解。本文提取了人体的运动轨迹特征、人体的运动速度特征、 人体的相对高度、人体的高宽比和人体区域的 Hu 矩特征，然后将其进行融合， 并利用决策树 SVM 作为分类器对人体的行为进行分类，取得了比较好的识别结 果。 关键词：行为分析 VIBE 算法 人体跟踪 遮挡处理 Abstract The video based human behavior analysis is the current research focus in computer vision. This paper has mainly studied three aspects which are respectively target detection, human body tracing and behavior description and understanding, and also developed a video based human behavior analysis system which can automatically detect and trace the human bodies (one or two) appearing in the video range, recognize each body’s simple behaviors and account each one’s sitting period. The main work of this article given as follows: Target detection.This paper has studied several common used target detection algorithms and the relatively novel VIBE algorithm, comparing them with each other. The experiments show that VIBE algorithm obtains better effect than others. Yet it cannot be used in the detection of quasi stationary targets. To solve this problem, this paper has improved VIBE algorithm combining with single Gauss model. At last, experimental comparison and analysis is made between the proposed algorithm of this paper and several common target detection algorithms and VIBE. At last, the results indicate that the proposed algorithm has better detecting effect than that of common used target detection algorithm and is almost equal with VIBE method; moreover, it can detect q