基于部件模型的攀爬行为检测算法研究

发布时间：2019-01-30 16:13

【摘要】：视频监控系统在公共安全领域发挥着日益重要的作用,随着城市各个公共场合安设的摄像头数量越来越多,视频的数据量也呈爆炸式增长,如何实现智能化监控成为当前研究的热点,而智能视频监控中研究的重点与难点就是人体行为的分析和识别,其应用十分广泛,如在人机交互和运动分析等领域。某些特定的行为如攀爬行为的检测同样具有广阔的应用前景,如小区,厂房,特别是在监狱,仓库等特殊场所。目前智能化的监控系统已经取得了一定的发展,但是由于人体姿态的多样性和拍摄背景的复杂性,使得针对人体的视频监控技术目前依然没有取得理想的结果。攀爬行为识别需要解决两个问题,一是检测问题,检测图片或者视频中是否有人体;二是识别问题,对人体的运动特征进行提取,通过对比模型识别出人体的行为。本文在这两点的基础上构建了人体攀爬识别系统,对人体的攀爬行为进行智能化的检测和分析,取得的主要成果如下:1.针对视频中人体的多样性问题,提出了一种基于可变形部件模型和颜色特征的多决策的人体检测方法,其基本思想是:首先使用可变形部件模型对图像进行初步的检测,然后判断检测的得分是否大于设定阈值,如果大于阈值且不在可疑区间,则判定为人体,如果在可疑区间则进一步对目标进行颜色特征检测并将检测的得分作为最终的判断结果,实验结果表明,基于可变形部件模型和颜色特征的多决策检测方法能够在一定程度上提高检测的准确率。2.利用可变形部件模型和密集轨迹特征,开发了一套人体攀爬检测系统,系统分为三大模块,分别是运动检测模块、人体检测模块和攀爬识别模块。检测流程是首先对视频中动态区域进行前景分离,提取需要检测的运动区域,然后使用可变形部件模型对运动区域进行人体检测,最后使用密集轨迹特征对人体的行为进行识别,如果是攀爬则进行标注或报警,达到智能化的人体攀爬监控目的。
[Abstract]:Video surveillance system is playing an increasingly important role in the field of public security. With the increasing number of cameras installed in various public places in cities, the amount of video data is also increasing explosively. How to realize intelligent surveillance has become a hot topic in current research, and the emphasis and difficulty of intelligent video surveillance is the analysis and recognition of human body behavior, which is widely used in the fields of human-computer interaction and motion analysis. The detection of certain behaviors such as climbing behavior also has a broad application prospect, such as community, factory building, especially in prison, warehouse and other special places. At present, intelligent monitoring system has made some progress, but due to the diversity of human posture and the complexity of shooting background, the video surveillance technology for human body has not yet achieved the desired results. There are two problems to be solved in the identification of climbing behavior, one is to detect whether there is a human body in a picture or video, the other is to identify the human body's behavior by extracting the motion characteristics of the human body. In this paper, a human climbing identification system is constructed on the basis of these two points, and the intelligent detection and analysis of human climbing behavior are carried out. The main achievements are as follows: 1. Aiming at the diversity of human body in video, a human body detection method based on deformable component model and color feature is proposed. The basic idea of the method is as follows: firstly, the deformable component model is used to detect the image. Then it is determined whether the test score is greater than the set threshold, and if the threshold is greater than the threshold and is not in a suspicious interval, then it is judged to be a human body, If the target is further detected by color features in the suspicious interval and the detected score is taken as the final judgment result, the experimental results show that, Multi-decision detection method based on deformable component model and color feature can improve the detection accuracy to some extent. 2. Based on the deformable component model and the dense track characteristics, a human climbing detection system is developed. The system is divided into three modules: motion detection module, human body detection module and climbing identification module. The detection process is to separate the foreground of the dynamic region in the video, extract the moving region to be detected, and then use the deformable component model to detect the moving region. Finally, the human behavior is identified by using the dense trajectory features, and if climbing, it is marked or alerted to achieve the purpose of intelligent human climbing monitoring.
【学位授予单位】：安徽大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.41;TN948.6

【相似文献】