基于背景差的运动目标检测方法比较分析*

背景差是常用的运动目标检测方法,其基本思想是通过视频帧与背景参考图像的差分实现运动目标检测。背景差法的核心是背景模型的构造。首先介绍基于背景差的运动目标检测方法的关键步骤:预处理、背景建模、背景差分、后处理;其次介绍几种典型的单背景和多背景模型;然后利用自适应背景模型、中值滤波、卡尔曼滤波、混合高斯模型等进行运动目标检测,并从速度、存储开销、准确率等方面对这些方法进行了分析比较。实验结果表明,单背景模型具有更快的检测速度,而多背景模型的检测准确度更高。

Comparison on Background Subtraction Algorithms for Moving Target Detection

Background subtraction is a typical approach to detection of moving targets,the idea of which is to subtract the current frame from a reference background image.To a large extent,the performance of moving target detection depends on construction of the background model.First,the key steps of moving targets detection is introduced,including pre-processing,background modeling,background subtraction and post-processing.Second,several classical mono-background and multi-background models are analyzed.Finally,adaptive background,median filter,kalman filter and mixture of Gaussians model are used to detect moving targets,and experimental results are given for comparing and analyzing these methods from the view points of speed,memory requirements and accuracy.Experimental results show the strongpoint of mono-background model is detection speed,but the multi-background models have more accuracy in results.