基于改进视觉背景提取的运动目标检测算法

发布时间：2018-02-25 06:12

本文关键词： 机器视觉运动目标检测视觉背景提取鬼影消除视觉显著性动态背景自适应阈值　出处：《光学学报》2016年06期 　论文类型：期刊论文

【摘要】：针对视觉背景提取算法(ViBe)中出现的鬼影问题、不能很好适应背景高频扰动和摄像机抖动问题以及由于采用空间邻域扩散机制引起背景更新错误问题,提出一种改进的视觉背景提取算法。该算法结合视觉显著性判断背景模型中存在的鬼影目标,通过判断背景模型中每个像素点的鬼影程度,结合模糊准则自适应改变时间子采样因子,加快消除鬼影的速度;通过建立一个闪烁程度矩阵,判断背景高频扰动程度来设置自适应匹配阈值,加入小目标丢弃和空洞填充策略;统计前景像素24邻域区域的像素点个数,判断前景像素点是否为摄像机抖动或者背景更新错误引起的噪点,提高算法的稳健性。结果表明,改进后的算法可以很好地弥补经典ViBe算法的不足,准确率与识别率等指标均大大提升。
[Abstract]:Aiming at the ghost problem in visual background extraction algorithm Vibe, it can not adapt to the background high frequency disturbance and camera jitter and background updating error due to the spatial neighborhood diffusion mechanism. An improved visual background extraction algorithm is proposed, which combines visual saliency to judge the ghost target in the background model, and determines the degree of ghost image of each pixel in the background model. Combined with fuzzy criterion, the time subsampling factor can be changed adaptively to speed up the elimination of ghost images, and the adaptive matching threshold can be set up by establishing a flicker degree matrix to judge the background high frequency disturbance degree. Adding small target dropping and empty filling strategy, counting the number of pixels in the foreground pixel 24 neighborhood, judging whether the foreground pixel is noise caused by camera jitter or background update error, and improving the robustness of the algorithm. The improved algorithm can make up for the shortcomings of the classical ViBe algorithm, and the accuracy and recognition rate are greatly improved.
【作者单位】：国防科学技术大学电子科学与工程学院;中国人民解放军61541部队;
【基金】：国家自然科学基金(61401474)
【分类号】：TP391.41

【参考文献】