基于达芬奇平台的视频异常事件检测算法研究与实现
发布时间：2018-09-18 10:52
【摘要】：视频监控摄像机的广泛使用和智能视频监控技术的发展带动了视频监控市场的蓬勃发展,以人作为视频监控主体的监控系统不再有能力实时处理由成百上千路摄像头全天候输入的海量监控视频。视频异常事件检测作为智能视频监控的重要分支,可以借助计算机视觉技术,从监控视频中主动检测出与大多数正常行为事件不相符合的少量异常行为事件,并及时发出报警信息,从而将传统的人从坐在屏幕前监控枯燥的工作中解脱出来。本文具体所做的工作有:1、分析A.Adam提出的基于观察点的异常事件检测算法原理及应用优缺点,针对其等间距观察点布置可能造成的不同环境下监控区域信息丢失及计算冗余,提出了基于场景的观察点自组织方案,在等间距观察点布置的基础上,实现不同监控场景观察点位置和密度的自动调整,应用性更强。2、基于SEED-DVS6446达芬奇开发板,实现了可运行在其DSP端的基于观察点的异常事件检测算法,以及ARM端的异常检测系统,最终形成了视频异常事件检测盒,接通电源后能够对接入的视频流实时检测是否发生异常事件并确定异常区域范围。3、使用混合高斯背景模型提取前景运动团块,光流法计算团块运动方向;分别采用“Hog+线性SVM”以及“Haar+级联结构AdaBoost”的方案在运动团块图像上进行行人和车辆检测;对检测到的行人或车辆采用团块跟踪获得其在视频场景中的运动轨迹。结合本文归纳的异常事件规则集,实现对监控场景下如人车越界、人车拌线等行为可描述的具体异常事件判别。从理论到实践,通过对前面三个部分内容的集成,实现了完整的视频异常事件管理系统:可以将达芬奇平台上基于观察点算法的异常事件检测系统,以及基于运动目标检测与跟踪的具体异常事件判别结合起来,协同运行,能有效地对视频场景中行人和车辆相关的已知和未知异常事件的实时检测。
[Abstract]:The wide use of video surveillance cameras and the development of intelligent video surveillance technology have led to the vigorous development of the video surveillance market. The surveillance system, which uses human as the main body of video surveillance, no longer has the ability to process the massive surveillance video which is input by hundreds of cameras all the time in real time. As an important branch of intelligent video surveillance, video abnormal event detection can actively detect a small number of abnormal behavior events from monitoring video, which does not accord with most normal behavior events. And timely send out alarm message, thus freeing the traditional people from sitting in front of the screen monitoring boring work. The specific work of this paper is to analyze the principle and application advantages and disadvantages of the anomaly event detection algorithm proposed by A.Adam based on observation points, aiming at the information loss and computational redundancy of monitoring area in different environments caused by the arrangement of observation points at equal spacing. Based on the arrangement of observation points at equal distance, the self-organization scheme of observation points based on scene is proposed. The automatic adjustment of observation points' position and density in different monitoring scenes is realized, and the application is stronger. 2. Based on SEED-DVS6446 da Vinci development board, An outlier event detection algorithm based on observation point and an anomaly detection system based on ARM are implemented in the DSP terminal. Finally, the video anomaly event detection box is formed. After the power supply is turned on, the abnormal event can be detected in real time and the range of abnormal area. 3. The foreground motion block can be extracted by using mixed Gao Si background model, and the moving direction of the block can be calculated by optical flow method. The schemes of "Hog linear SVM" and "Haar cascaded structure AdaBoost" are used to detect pedestrian and vehicle on the motion cluster image, respectively, and the detected pedestrian or vehicle is tracked by cluster block to obtain their motion track in the video scene. Combined with the rule set of abnormal events summarized in this paper, we can distinguish the concrete abnormal events which can be described under the monitoring scenario such as the behavior of the human vehicle crossing the boundary, the human-vehicle mixing line, and so on. From theory to practice, through the integration of the first three parts, a complete video anomaly event management system is implemented: the anomaly event detection system based on the observation point algorithm can be used on the Da Vinci platform. By combining the detection of moving targets with the discrimination of specific abnormal events, the real-time detection of known and unknown abnormal events related to pedestrians and vehicles in the video scene can be effectively realized.
【学位授予单位】：电子科技大学
【学位级别】：硕士
【学位授予年份】：2015
【分类号】：TN948.6
本文编号：2247680
[Abstract]:The wide use of video surveillance cameras and the development of intelligent video surveillance technology have led to the vigorous development of the video surveillance market. The surveillance system, which uses human as the main body of video surveillance, no longer has the ability to process the massive surveillance video which is input by hundreds of cameras all the time in real time. As an important branch of intelligent video surveillance, video abnormal event detection can actively detect a small number of abnormal behavior events from monitoring video, which does not accord with most normal behavior events. And timely send out alarm message, thus freeing the traditional people from sitting in front of the screen monitoring boring work. The specific work of this paper is to analyze the principle and application advantages and disadvantages of the anomaly event detection algorithm proposed by A.Adam based on observation points, aiming at the information loss and computational redundancy of monitoring area in different environments caused by the arrangement of observation points at equal spacing. Based on the arrangement of observation points at equal distance, the self-organization scheme of observation points based on scene is proposed. The automatic adjustment of observation points' position and density in different monitoring scenes is realized, and the application is stronger. 2. Based on SEED-DVS6446 da Vinci development board, An outlier event detection algorithm based on observation point and an anomaly detection system based on ARM are implemented in the DSP terminal. Finally, the video anomaly event detection box is formed. After the power supply is turned on, the abnormal event can be detected in real time and the range of abnormal area. 3. The foreground motion block can be extracted by using mixed Gao Si background model, and the moving direction of the block can be calculated by optical flow method. The schemes of "Hog linear SVM" and "Haar cascaded structure AdaBoost" are used to detect pedestrian and vehicle on the motion cluster image, respectively, and the detected pedestrian or vehicle is tracked by cluster block to obtain their motion track in the video scene. Combined with the rule set of abnormal events summarized in this paper, we can distinguish the concrete abnormal events which can be described under the monitoring scenario such as the behavior of the human vehicle crossing the boundary, the human-vehicle mixing line, and so on. From theory to practice, through the integration of the first three parts, a complete video anomaly event management system is implemented: the anomaly event detection system based on the observation point algorithm can be used on the Da Vinci platform. By combining the detection of moving targets with the discrimination of specific abnormal events, the real-time detection of known and unknown abnormal events related to pedestrians and vehicles in the video scene can be effectively realized.
【学位授予单位】：电子科技大学
【学位级别】：硕士
【学位授予年份】：2015
【分类号】：TN948.6
【共引文献】
相关期刊论文 前4条
1 王卫;金立左;;基于DSP+FPGA视觉跟踪平台的设计与实现[J];工业控制计算机;2015年03期
2 袁承宗;李贤;张震;叶盛;蔡非凡;;基于以太网络的宇航用高清摄像机的硬件设计[J];电视技术;2015年11期
3 孙晓康;;基于DM6446的高效Linux启动方案设计[J];信息技术;2014年08期
4 王洪伟;;基于DM3730和GStreamer的网络视频传输技术设计与实现[J];应用科技;2014年03期
相关博士学位论文 前1条
1 陈龙;辅助视觉中的图像处理关键技术研究[D];西安电子科技大学;2013年
相关硕士学位论文 前10条
1 朱道广;高清CCD网络相机后端视频处理系统设计与实现[D];南京理工大学;2014年
2 赵孝磊;基于DM368高清智能监控系统的设计[D];南京理工大学;2014年
3 刘法龙;面向图像融合应用的红外/SAR图像仿真、采集、传输技术研究[D];杭州电子科技大学;2014年
4 郑亮;基于DM8168多路视频监控系统研制[D];杭州电子科技大学;2014年
5 张震震;基于DSP的视频监控系统的设计[D];太原科技大学;2014年
6 潘明;基于H.264的网络视频监控系统的研究与实现[D];吉林大学;2014年
7 丁盼盼;基于视频技术的车辆检测与跟踪算法的研究与实现[D];扬州大学;2013年
8 肖灵通;基于Davinci的便携式热成像检测终端设计与实现[D];南京航空航天大学;2014年
9 陈立峰;基于H.264的码率控制算法研究与编码器优化[D];中国计量学院;2014年
10 杨宗旭;基于达芬奇平台的机械臂视觉伺服系统设计[D];哈尔滨工业大学;2014年
，本文编号：2247680
本文链接：https://www.wllwen.com/kejilunwen/wltx/2247680.html
