摘要
树叶晃动、光照变化等自然场景下的动态背景会影响运动目标检测的准确性,区分动态背景和前景目标的变化是复杂场景下运动目标检测的首要任务。针对现有的前景提取算法逐点提取前景从而导致计算资源浪费的问题,提出了一种区域提取与改进LBP(Local Binary Patterns)纹理特征相结合的运动目标检测算法。首先,将图像分为大小相等的图像块,利用各图像块的统计特性建立核密度估计(Kernel Density Estimation,KDE)模型,并用KDE模型估计出前景区域。然后,计算前景块中所有像素点的改进LBP纹理特征直方图。最后,通过直方图匹配提取所有的前景像素实现目标的精确提取,并用概率模型更新背景。实验结果表明,该方法在快速提取运动目标前景区域的同时能够消除大部分动态背景产生的干扰,相比传统算法更适用于自然场景下的运动目标检测。
Detection accuracy of moving object is dramatically affected by the dynamic natural background,for instance,the shaking leaves and varying illumination.Therefore,it is essential to distinguish between the dynamic background and the foreground moving object.The existing foreground extraction algorithm extracts the foreground point by point,which leads to a waste of computing resources.This paper proposed a novel moving object detection algorithm based on region extraction and improved Local Binary Patterns(LBP).First,the image is divided into several image blocks of same size,and the Kernel Density Estimation(KDE)model is established according to the statistical characteristics of these image blocks.The foreground region is estimated by the KDE model.Then,the improved LBP texture feature histogram of all pixels in the foreground block is obtained.By ma-tching the histogram,all the foreground pixels are extracted,and the background is updated with a probabilistic model.The experimental results show that the proposed method can quickly extract the foreground region of moving target and eliminate most of the interference caused by dynamic background.Compared with the traditional algorithm,the proposed method is more suitable for moving object detection in natural scenes.
作者
辛元雪
史朋飞
薛瑞阳
XIN Yuan-xue;SHI Peng-fei;XUE Rui-yang(College of Internet of Things Engineering,Hohai University,Changzhou,Jiangsu 213022,China)
出处
《计算机科学》
CSCD
北大核心
2021年第7期233-237,共5页
Computer Science
基金
国家自然科学基金(61801168,61801169)
江苏省自然科学基金(BK20170305)。
关键词
运动目标检测
动态背景
KDE
区域提取
LBP纹理特征
Moving object detection
Dynamic background
Kernel density estimation
Region extraction
Local binary patterns texture features
