基于光流特征的群体异常行为检测方法的研究
发布时间：2018-10-22 06:48
【摘要】：随着社会经济的不断发展,人群密集的公共场所越来越多,如何对人群进行有效的监控已经成为公共安全中的突出问题。智能视频监控采用计算机视觉、图像处理和模式识别等技术对人群状态进行监控,可以有效的检测到人群中的异常行为。本文在研究分析群体异常行为检测相关技术基础上,设计了一种基于光流特征的群体异常行为检测方法。本文首先对现有的运动目标检测方法进行了研究分析,在检测运动目标时使用了混合高斯模型的方法,并对检测到的目标区域使用中值滤波器进行平滑去噪处理,可以得到连通性和准确度更为理想的运动目标区域。在对目标特征进行提取时,本文在研究分析Harris角点提取算法的基础上,设计了一种改进的多尺度Harris角点提取算法,使用该方法提取的特征点在数量上更加充足而且性质更加稳定。其次,在进行群体异常检测时,本文先使用金字塔Lucas-Kanade光流法对提取的特征点进行跟踪匹配,进而可以得到群体的光流场,通过光流场中的光流矢量信息提取群体的运动特征,本文提取的运动特征包括运动的平均动能和方向熵,将计算得到的运动特征值与预先设定的阈值进行比较来判断群体是否存在异常,在进行比较判断时本文采用连续五帧的运动特征值与阈值进行比较,如果连续五帧的值都大于阈值,则判断群体存在异常行为。最后,为了验证本文设计的群体异常检测方法性能,通过对UMN数据集进行实验测试,实验结果表明本文设计的群体异常检测方法在准确率和实时性方面都比较理想。
[Abstract]:With the continuous development of social economy, more and more crowded public places, how to effectively monitor the crowd has become a prominent problem in public security. Intelligent video surveillance uses computer vision, image processing and pattern recognition technology to monitor the state of the crowd, can effectively detect abnormal behavior in the crowd. In this paper, based on the research and analysis of the correlation technology of group abnormal behavior detection, a method based on optical flow characteristics is designed to detect group abnormal behavior. In this paper, the existing methods of moving target detection are studied and analyzed, and the mixed Gao Si model is used to detect the moving target, and the median filter is used to smooth the target region. A moving target region with better connectivity and accuracy can be obtained. On the basis of studying and analyzing the Harris corner extraction algorithm, an improved multi-scale Harris corner extraction algorithm is designed in this paper. The feature points extracted by this method are more abundant in quantity and more stable in nature. Secondly, in the process of group anomaly detection, the pyramidal Lucas-Kanade optical flow method is used to track and match the extracted feature points, and then the optical flow field of the group can be obtained, and the motion characteristics of the group can be extracted by the optical flow vector information in the optical flow field. The motion features extracted in this paper include the average kinetic energy and directional entropy of motion. The calculated motion eigenvalues are compared with the pre-set threshold to determine whether the population is abnormal or not. In this paper, the motion eigenvalues of five consecutive frames are compared with the threshold. If the values of the five consecutive frames are all larger than the threshold, the group has abnormal behavior. Finally, in order to verify the performance of the group anomaly detection method designed in this paper, the experimental results of UMN data set show that the proposed method is ideal in both accuracy and real-time.
【学位授予单位】：西安理工大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.41;TN948.6
本文编号：2286405
[Abstract]:With the continuous development of social economy, more and more crowded public places, how to effectively monitor the crowd has become a prominent problem in public security. Intelligent video surveillance uses computer vision, image processing and pattern recognition technology to monitor the state of the crowd, can effectively detect abnormal behavior in the crowd. In this paper, based on the research and analysis of the correlation technology of group abnormal behavior detection, a method based on optical flow characteristics is designed to detect group abnormal behavior. In this paper, the existing methods of moving target detection are studied and analyzed, and the mixed Gao Si model is used to detect the moving target, and the median filter is used to smooth the target region. A moving target region with better connectivity and accuracy can be obtained. On the basis of studying and analyzing the Harris corner extraction algorithm, an improved multi-scale Harris corner extraction algorithm is designed in this paper. The feature points extracted by this method are more abundant in quantity and more stable in nature. Secondly, in the process of group anomaly detection, the pyramidal Lucas-Kanade optical flow method is used to track and match the extracted feature points, and then the optical flow field of the group can be obtained, and the motion characteristics of the group can be extracted by the optical flow vector information in the optical flow field. The motion features extracted in this paper include the average kinetic energy and directional entropy of motion. The calculated motion eigenvalues are compared with the pre-set threshold to determine whether the population is abnormal or not. In this paper, the motion eigenvalues of five consecutive frames are compared with the threshold. If the values of the five consecutive frames are all larger than the threshold, the group has abnormal behavior. Finally, in order to verify the performance of the group anomaly detection method designed in this paper, the experimental results of UMN data set show that the proposed method is ideal in both accuracy and real-time.
【学位授予单位】：西安理工大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.41;TN948.6
【参考文献】
相关期刊论文 前2条
1 王尔丹;人群运动与密度估计技术研究[J];安全;2005年03期
2 庄越挺,傅正钢,叶朝阳,吴飞;基于视听分层模型的实时爆炸场景识别[J];计算机辅助设计与图形学学报;2004年01期
相关硕士学位论文 前1条
1 符祖会;智能监控系统中行为识别关键技术研究与实现[D];电子科技大学;2013年
，本文编号：2286405
本文链接：https://www.wllwen.com/kejilunwen/xinxigongchenglunwen/2286405.html
