【作者】 张旭；

【导师】 万丽莉；

【作者基本信息】 北京交通大学 ， 信号与信息处理， 2012， 硕士

【摘要】 摘要：随着计算机硬件水平以及计算机视觉学科的发展,智能视频监控越来越受到研究机构和学者的关注,一些智能视频监控的产品也逐渐出现。当前的智能视频监控主要是基于目标检测与跟踪的解决方案,其智能性还不够高,本文针对人们对智能视频监控不断增长的需求,提出了一种基于动作识别的智能视频监控解决方案。本文分为人体检测、人体跟踪以及动作识别三个部分进行详细论述。由于人体检测与跟踪的成熟性,本文的重点放在了动作识别相关的研究上。在实现基本人体检测与跟踪的算法基础上,本文实现了基于投影直方图的动作识别与基于词袋模型和主题模型的动作识别,并对后者进行了改进,提出了一种基于光流描述子词袋的动作识别方法。由于前人在构建词库时,对各类动作进行单词频率统计时,总是按照全局来统计的,这样可能会导致单词的误匹配,为了解决这个问题,我们利用了人体模型的特点,认为动作在时间轴上独立,人体各个部分做的动作在空间上是相关的。我们以光流局部时空最大值作为特征,将人体区域分块构建光流描述子词库,强制将人体不同部位提取到的相似特征分配为不同的单词,从而在词库中加入空间信息。最后,我们改进了原有的pLSA模型,并使用改进的pLSA模型进行动作识别。实验证明,该算法在Weizmann视频数据库与KTH视频数据库上有着良好的性能。为了验证算法在实际环境的有效性,本文搭建了一个智能视频监控平台,平台具有基本的检测、跟踪算法,基于光流描述子词袋的动作识别算法和基于投影直方图的动作识别算法等,并可以通过短信发送模块将监控信息及时发送到指定手机,记录检测、跟踪以及识别的结果。和普通智能监控平台相比,我们的监控平台更加智能有效。更多还原

【Abstract】 The intelligent visual surveillance has attracted significant interest in recent years. Some intelligent visual surveillance product has appeared in public areas as well. However most of the products are based on motion object detecting and tracking. They are not intelligent enough. In this paper, we propose an intelligent visual surveillance solutions based on the human action recognition.There are three important approaches, human detection, tracking and action classifying, in human action recognition. Recently, the approaches of human detection and tracking are efficient enough. So we focus on the approach of human action recognition classifying. We complete the basic algorithm of the human detection and tracking and present an improved approach to classify human action based on the BOW model and the pLSA (probabilistic Latent Semantic Analysis) model. We propose an improved feature, which is called the local spatial-temporal maximum value of optical flow to build our bag of words. This feature is able to reduce the high dimension of the pure optical flow template and also has abundant motion information. Our approach of recognition is tested on two datasets, the KTH datasets and WEIZMANN datasets. The result shows its good performance.In order to test our method in reality, we develop an intelligent visual surveillance platform. This platform has basic function of human detection and tracking. The human action classifying algorithm based on our improved approach and a method based on the projection histogram are also in the platform. The surveillance information can be sent to the mobile phone automatically. Compared to the common intelligent visual surveillance platform, our intelligent visual surveillance platform is more intelligent and effective.更多还原