基于全方位计算机视觉的盗窃事件检测

Theft detection based on omni-directional vision sensors

为了实现公共场所的安防监控智能化,结合全方位视觉传感器(ODVS)、动态图像处理等技术设计出一种盗窃事件检测系统。首先,通过ODVS来获得360°无死角、大范围的全景视频防盗检测区域;其次,提出了一种基于两个不同画面更新率的混合高斯模型进行背景建模的动态图像处理方法来获取特殊背景对象,同时还能区分场景内的运动对象和纯背景对象;将被盗窃的物体作为特殊背景对象进行检测。实验结果表明,该盗窃事件检测系统具有检测范围广、检测精度高、鲁棒性好和实时性高等优点。

This paper designed a system about theft detection to realize smart video surveillance in public locations. Firstly, aiming at the problem of small visual field, Omni-Directional Vision Sensors （ODVS） which have 360 degree and nondead-angle view were used to capture panoramic images of scene. Secondly, by processing the input video at different frame rates, two backgrounds were constructed： one for short-term and the other for long-term. A special background was detected by comparing the current frame with the background. Meanwhile, moving objects and static background could also be distinguished. The stolen object would be detected as a special background. Experimental results show that the system is robust enough to detect the stolen objects effectively.