基于Kinect的室内异常行为检测

发布时间：2018-10-30 20:28

【摘要】：近几十年来,科学技术不断地进步和提高,视频监控尤其是高清系统也得到了普及,计算机视觉处理技术也得到了的提升。使用计算机视觉处理技术对HD视频进行处理,将计算机视觉处理技术应用到安防监控领域,以提高公共场合的安全性,主要是通过检测公共场合出现的异常行为并对出现的异常行为向人们报警。计算机视觉技术广泛地应用到视频监控领域,同时也需要高效率的算法去解决实时问题。在计算机视觉领域中,科研工作者不断地用计算机对“人体行为”进行识别和理解,从前景目标检测、目标的追踪定位和最终对行为进行理解。采集的视频由于受到光照、阴影、遮挡和噪声等影响,行为的理解也存在一定的困难。由于Kinect的出现,使得深度图像(RGB-D)进入人们的关注领域,Kinect传感器受到的外界干扰小,在黑暗的环境下也可以识别目标人体,可以获取骨骼特征,具有空间特性。可用于人体行为识别,由此科研工作者引发了高度兴趣并激发新的灵感和解决思路,基于Kinect平台对异常行为进行检测。本文就是使用Kinect设备检测室内的异常行为,RGB-D就是获取的数据信息。论文中研究的异常行为是针对室内中的场景,对这些不符合人们预期的行为进行检测,通常异常行为包括:跌倒,打斗,追逐等等。并对检测到的异常进行报警。本文首先叙述了人体异常行为检测的三个阶段所使用的算法和特征,分析这些算法和特征的优缺点,讨论了研究现状和面临的问题和难点,并分析了使用图片深度信息和骨架关节点信息的可行性。其次介绍了Kinect硬件设备、软件架构,阐述了它是如何获取RGB-D信息的,接着对骨架关节点进行了描述。并对采集的数据提取骨架关节点信息,使用关节点角度信息进行特征表示,使用这些特征对行为进行区分。接着,本文对主流的人体行为识别算法进行介绍,本文采用动态规整算法对人体行为进行检测,并对该算法进行改进,提高运行效率。最后,论文概括了研究工作,然后对即将开展的工作和发展趋势进行了讨论和展望。
[Abstract]:In recent decades, with the continuous progress and improvement of science and technology, video surveillance, especially high-definition system has been popularized, computer vision processing technology has also been improved. Computer vision processing technology is used to process HD video, and computer vision processing technology is applied to the field of security monitoring, in order to improve the security of public places. It mainly detects abnormal behavior in public and alerts people about abnormal behavior. Computer vision technology is widely used in the field of video surveillance. At the same time, it needs efficient algorithms to solve real-time problems. In the field of computer vision, researchers constantly use computers to identify and understand "human behavior", from foreground target detection, target tracking and orientation, and ultimately to understand behavior. Because of the influence of illumination, shadow, occlusion and noise, it is difficult to understand the behavior of the collected video. Because of the appearance of Kinect, the depth image (RGB-D) has come into the field of attention. The Kinect sensor has little external interference, and can recognize the target human body in the dark environment. It can obtain the bone feature and have the spatial characteristic. It can be used to identify human behavior, which arouses high interest, inspiration and solution, and detects abnormal behavior based on Kinect platform. This article uses the Kinect device to detect the abnormal behavior in the room, and RGB-D is the data information obtained. The abnormal behaviors studied in this paper are aimed at indoor scenes and detect these behaviors which do not meet the expectations of people. The abnormal behaviors usually include falling, fighting, chasing and so on. And the detection of abnormal alarm. This paper first describes the algorithms and features used in the three stages of human abnormal behavior detection, analyzes the advantages and disadvantages of these algorithms and features, and discusses the present research situation, problems and difficulties. The feasibility of using image depth information and skeleton node information is analyzed. Secondly, the hardware and software architecture of Kinect are introduced, and how to obtain RGB-D information is described. Then the skeleton node is described. The skeleton node information is extracted from the collected data, and the angle information of the node is used to represent the feature, and the behavior is distinguished by these features. Then, this paper introduces the mainstream human behavior recognition algorithm, this paper uses dynamic warping algorithm to detect human behavior, and improves the algorithm to improve the running efficiency. Finally, the research work is summarized, and the future work and development trend are discussed and prospected.
【学位授予单位】：吉林大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.41

【相似文献】