基于三角核估计模型的运动目标快速检测方法

在资源受限情况下,针对视频监控系统的实时性要求,在摄像头静止场景下,通过三角核估计法并结合改进的帧差法实现运动目标检测.首先通过帧差法对场景中物体运动进行判断,并结合背景减法对于运动点进行预判,对已判为预运动点部分采用三角核估计模型对目标进行精确提取,在保证精度的情况下显著减少了运算量.同时引入一种快速有效的背景模型更新模式,使三角核估计法的样本数目大大减少,节约了硬件资源,提高了系统性能.

A Motion Detection Approach Based on Context Modeling

In allusion to the real time requirement of video surveillance system, with static camera in this paper, triangle kernel estimation and an improved frame-difference model is used to model the scene background. At first, the frame-difference model is used to detect whether there is something moving. If motion is detected, background subtraction model is used to detect the moving pixel. In the end, we use kernel density estimation to strictly determine the moving pixel. This model is faster than the original kernel density estimation mode with same accuracy. Also this paper diverts to a background update model. It cuts down the sample quantity effectively.