基于结构相似性粗定位与背景差分细分割的运动目标检测方法  

Moving Target Detection Method Based on Rough Positioning of Structure Similarity and Fine Segmentation of Background Difference

作　　者：蒙晓宇 朱磊[1] 张博[1] 潘杨[1] MENG Xiao-yu;ZHU Lei;ZHANG Bo;PAN Yang(College of Electronics and Information, Xi'an Polytechnic University, Xi'an 710048, China)

机构地区：[1]西安工程大学电子信息学院,西安710048

出　　处：《科学技术与工程》2021年第36期15563-15570,共8页Science Technology and Engineering

基　　金：国家自然科学基金(61971339);陕西省科技厅重点研发计划(2019GY-113)。

摘　　要：针对抖动相机和静止相机下的运动目标检测问题,提出基于结构相似性粗定位与背景差分细分割的运动目标检测方法。首先使用动态模式分解法根据视频序列提取彩色背景图像为粗定位提供基础,提出在小范围内利用相关法对尺度不变特征变换(scale-invariant feature transform,SIFT)算子检测到的当前帧图像和彩色背景图像的特征点进行匹配,通过匹配点对的偏移量估计当前帧图像的偏移程度,以达到消除图像抖动的目的;然后利用结构相似性对目标区域粗定位,减少复杂背景的干扰;再对各通道下粗定位彩色背景图像及校正后的当前帧图像背景差分并对其结果进行与操作;最后通过形态学处理得到完整的运动目标。结果表明:所提方法不仅有效改善了相机的抖动问题,而且在抖动相机和静止相机两种情况下的检测率有所提高,与混合高斯模型(Gaussian mixture model,GMM)等3种算法相比查全率和准确率分别提高1.6%、3.5%和3%以上。Aiming at the problem of moving object detection in dither camera and still camera,a moving target detection method based on rough location of structure similarity and fine segmentation of background difference was proposed.Firstly,the dynamic mode decomposition method was used to extract the color background image from the video sequence to provide the basis for coarse positioning.In a small range,the correlation method was used to match the feature points of the current frame image and the color background image was detected by the scale-invariant feature transform(SIFT)operator,the offset degree of the current frame image was estimated by matching the offset of the point pair,in order to achieve the purpose of eliminating image jitter.Then the structural similarity was used to coarsely locate the target area to reduce the interference of complex background.Then the coarse positioning color background image under each channel and the corrected background difference of the current frame image were compared,and the results were analyzed and operated.Finally,the complete movement target was obtained through morphological processing.The results show that the proposed method not only effectively improves the problem of camera shake,but also improves the detection rate in both cases of shaking camera and still camera.Compared with the three algorithms such as Gaussian mixture model(GMM),the recall rate,recall rate and accuracy rate are increased by at least 1.6%,3.5%and 3%.

关 键 词：运动目标检测 特征匹配 结构相似性 前景偏移估计 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]