基于运动轨迹的视频语义事件建模方法

Modeling approach of the video semantic events based on motion trajectories

人体行为分析是计算机视觉领域的重要课题之一。针对人体行为事件,提出了一种基于运动轨迹的视频语义事件建模方法。首先,采用改进的基于Surendra背景建模算法检测运动行人目标,然后利用Meanshift跟踪算法得到目标行人的运动轨迹路径,最后根据人体行走轨迹特征和所定义语义事件模型进行相关事件判断,并搭建平台实现视频语义事件自动监测。对监控视频公开数据集的实验测试表明,提出的方法可准确有效的识别常见的人体行为,为视频语义领域提供一种可靠、准确的技术方案。

The analysis of human behavior is one of the important topics in the field of computer vision. Aiming at human behavior events, this paper presents a modeling approach of the video semantic events based on motion trajectories. Firstly, we adopt the improved Surendra background modeling algorithm to detect moving targets. And then,by making use of the Meanshift algorithm to get the trajectories of the pedestrians, we can judge these related events according to human walking trajectory features and the defined modeling approach of the video semantic events. After that,a system is built to achieve automatic monitoring of video semantic events. The test result of surveillance video public data sets shows that the proposed method can recognize common human behaviors accurately and effectively,and provides one of the reliable and accurate technical solutions in video semantic field.