视频流中人体的检测与分离.pdfVIP

摘要 视频流中人体的检测与分离已经成为计算机视觉重要的研究热点。它在智能 视频监控以及图像储存与检索等领域具有广阔的应用前景。本文从人体检测与多 个人体分离两个方面进行了研究，主要内容是使计算机从单目视频中捕获完整的 前景目标后，提取人体特征，通过模式识别来识别人体。 在人体检测部分，首先对视频序列中视觉场景进行建模，在利用背景减除法 将运动目标与背景分离后，然后对运动目标进行二值化。由于目标对象存在阴影 区域，所以接着需要对影子进行滤除。再通过中值滤波以及 Canny 算法提取运动 对象的轮廓，从而得到了完整的轮廓，最后提取其人体特征。在特征提取算法中， 本文分析了传统人体特征提取算法在特征点上计算的复杂性，提出了基于像素点 的人体特征算法,该算法对遮挡问题有较强的适用能力。在分类识别部分，针对于 人体与非人体两类样本分类问题，采用支持向量机作为分类器。 在特定人体的分离上，本文在视频序列中提取人体上 SURF 特征点。分离不 同人体上特征点，利用零均值归一化算法对前后两帧进行匹配，经过扩散后达到 了分离目的。 采集室内外各种背景下人体运动视频序列，利用 VisionLab 平台经过多次实验, 本文研究方法具有广泛的实用性。 关键词： 人体检测 识别 人体特征 人体分离 SURF 算法 Abstract The detection and separation of human body in video stream is becoming a hotspot topic in the domain of computer vision reseach. This strong interest is driven by a wide spectrum of promising applications in areas such as smart surveillance, image storage and retrieval, etc. This paper has done some research on detection and separation of multiply human bodys individual. The mainly research work is to capture complete foreground, extract human body features, and recognize normal human body by pattern recognition from single video image. In human body detection part, before detecting moving objects, we need “Background Modeling” to the visual scenes. The binary maps are captured from the binaryzation to Background Subtraction process. Unfortunately the binary maps maybe have shadows while the objects are moving, so we must move these shadows from them, and we can get complete contours behind median filter and canny algorithm, then we extract human body features. As far as the human body feature exaction is concerned, traditional algorithm is of much computation complexity.A pixel-based feature extraction algorithm of human shape for classification is proposed in th