【作者】 周蓉君；

【导师】 李斌；

【作者基本信息】 中国科学技术大学 ， 电路与系统， 2010， 硕士

【摘要】 人体检测和跟踪是计算机视觉的重要研究课题之一,其应用已经遍及智能监控、体育运动分析、新型人机交互和虚拟现实等领域,研究基于视觉的人体检测和跟踪有着很重要的现实意义。本文研究了在智能家居环境中人体检测与人体运动跟踪问题,主要工作和特色如下:1.对人体运动图像分析的关键技术进行了综述。详细介绍了人体运动图像中从初始化模型、人体跟踪、姿态估计和行为识别的各个步骤,及其主要的研究方法与进展,以及存在的问题。2.研究了一种改进的自适应混合高斯模型的人体检测算法。家居背景经常发生变化,需要采用混合高斯函数来做背景建模。已有的自适应混合高斯模型可以快速对背景进行建模,但是当人体在某个空间位置短暂停留时,就会被学习成背景。本文采用一种近似S曲线函数的分段指数函数来修正学习率和新的背景高斯成分的权重,并结合形态学、中值滤波和图像金字塔技术,有效地减除了人体阴影,降低了背景噪声的影响。3.研究了一种融合了混合高斯模型和粒子群优化的自动人体跟踪算法利用混合高斯背景模型给出的目标前景区域,缩小跟踪的范围;利用粒子群优化方法对状态空间的随机搜索能力,寻找人体目标的最优空间位置,最后将两者的信息融合对比,实现自动的人体跟踪算法。本算法框架为每个跟踪目标维护一个历史灰度直方图信息,并在人体遮挡、人体目标离开场景等目标消失情况下,可以在稍后图像帧中自动地恢复跟踪和识别目标。更多还原

【Abstract】 The detection and tracking of human body is an important problem in the computer vision field. Its application has spread intelligent monitoring, sports analysis, a new human computer interaction and virtual reality and other fields. Researching the technology of the human detection and tracking based on vision has very important practical significance. Body detection in the smart home environment and human motion tracking based on video have been studied in this dissertation. The main work and characteristic are as fol-lows:1. The key technologies of human movement image analysis were reviewed.This paper gives detail steps in the human motion image which contain the initialization model, human tracking, pose estimation and behavior recognition, the major research methods, the progress and problems.2. Research an improved adaptive Gaussian mixture model for human detection algo-rithm.The frequent changes in the home background need to be described by the mixed Gaussian function modeling. Existing adaptive Gaussian mixture models can quickly model the background, but when the body stays a short time at a position, they will learn it for the background. S-Curve function which is an approximate piecewise exponential function is used to modify the background learning rate and the new Gaussian component weights. After that, a combination of morphology, median filtering and image Pyramid Technology, is used to cuts off human shadow and reduces the background noise.3. Research an automatic human tracking algorithm by the fusion of Gaussian mixture model and particle swarm optimization.The target foreground area given by Gaussian mixture background model is used to narrow the range of human tracking. The random search capability of particle swarm op-timization method in the state space is used to find the optimal spatial location of the body. Finally, the information fusion between them can achieve automatic human tracking al-gorithm. The algorithm framework maintains a history of histogram information for each tracking object, and can automatically recover tracking and identify targets in the later image frame when self-occlusion or human target leaving the scene.更多还原