【作者】 姜楠；

【导师】 牛连强；

【作者基本信息】 沈阳工业大学 ， 计算机应用技术， 2009， 硕士

【摘要】 行人检测是智能车辆辅助导航系统中的一项关键技术,也是目前计算机应用领域的研究热点之一。它处于智能车辆辅助导航系统的底层,是各种后续高级处理如目标分类、行为理解的基础。行人检测技术在智能控制系统、虚拟现实、机器人应用等方面也将得到广泛地应用。论文对车辆辅助驾驶系统中基于计算机视觉的行人检测技术进行研究,首先详细分析了基于视频图像的运动目标检测方法,重点研究了减背景法,提出了背景建模方法的改进算法,该算法是借助图像中相邻象素的邻域相关性,只对间隔的象素点建立混合高斯模型,没有建立混合高斯模型的点则利用象素的邻域相关性进行判断;为了克服了减背景法对光照变化敏感的不足,再结合帧差法对运动目标进行检测。其次,对运动目标进行行人识别,根据行人多种简单特征建立行人匹配模板,在图像序列中对检测到的目标进行匹配和识别,实时的分割出准确完整的运动行人。最后,对行人运动跟踪技术进行了研究,实现了一种以运动目标质心模型求得运动目标位置的测量方法,并结合卡尔曼滤波方法对行人的矩形窗口进行预测,建立动态可变的感兴趣区域,缩小了行人的搜索范围。论文采用手持摄像机的形式来模拟装载在智能车辆上的可变焦距的摄像机,对采集到的视频图像进行处理。本文提出的算法能够实现对车辆前方的行人进行实时有效地检测识别,同时对行人进行跟踪。为了验证算法的有效性,进行了一定量的试验,试验结果表明,该算法可以满足行人检测系统的实时性要求,相对于以往算法也更加准确可靠。更多还原

【Abstract】 The technique of moving pedestrian detection is one of the key techniques for Intelligent Vehicle Navigation System and is coming to a hot spot of research in computer application field at present. It is in the Intelligent Vehicle assisted navigation technology bottom, splitting the moving objects real-time and effectively is essential foundation for the subsequent processing, including target identification, tracking behavior comprehension. Method of moving pedestrian detection will also be widely used in the intelligent control system, virtual reality, robot applications.This paper is study on the Computer Vision Based Pedestrian Detection for Driver Assistance Systems. First of all, a detailed analysis of video images based on the moving target detection methods, the dissertation mainly studied the background subtraction, and improved background modeling method, considering the pixels neighboring relativity, the algorithm through the interval pixels establishment of Gaussian mixture model instead of the traditional point by point establishes Gaussian mixture model in background subtraction. In order to overcome the lack of sensitivity of light change, combining the adaptive background subtraction with the symmetrical differencing obtains the integrity foreground image. Then, during the process of matching, the template for pedestrian matching and recognizing in image sequence was presented by fusing several characters of targets. Finally, on the basis of moving pedestrian segmentation in this paper, continue to process the Video images, combined with Kalman filter estimates the position of the pedestrian bounding box, A dynamic and variable area of interest is formed to narrow the seek range of pedestrian .In the experiments, used hand-held video camera to replace the video camera equipped on the moving vehicle to process the images. The experiment results indicate that the algorithms introduced could achieve effective recognition of the proceeding pedestrians. At the same time, the algorithms have good tracking. A serial of experiment have been implemented to verify the effectiveness of the method supported by the papers, which can satisfied with the requisition of the real time system and have been more precisely.更多还原