融合MOG2算法与FT算法的移动目标检测方法  

A Moving Target Detection Method Combining MOG2 Algorithm and FT Algorithm

作　　者：刘泽平 刘明兴 李晓帆 姚明杰 麻方达 符朝兴[1] LIU Zeping;LIU Mingxing;LI Xiaofan;YAO Mingjie;MA Fangda;FU Chaoxing(College of Mechanical and Electrical Engineering,Qingdao University,Qingdao 266071,China)

机构地区：[1]青岛大学机电工程学院,山东青岛266071

出　　处：《青岛大学学报（工程技术版）》2022年第2期61-66,共6页Journal of Qingdao University(Engineering & Technology Edition)

摘　　要：针对移动目标检测方法存在的受背景环境影响而导致检测结果不准确的问题,本文提出一种融合高斯混合模型的背景减除法。算法与频率调谐显著性区域检测(frequency-tuned,FT)算法的移动目标检测方法。该研究以摄像头拍摄到的遥控车行走视频为分析材料,在降采样图像上用MOG2算法检测移动目标,经二值化和滤波去噪后框选移动目标感兴趣区域,采用FT算法检测此感兴趣区域,并利用置信度结合两次检测结果完成移动目标检测。实验结果显示,在置信度分别为α=0.6,β=0.4时,本文检测方法将检测300帧图像时平均交并比由0.72提高到0.78,说明该算法可以消除背景阴影及鬼影影响,从而有效提高移动目标的检测精度。The moving target detection method is affected by the background environment,resulting in inaccurate detection results.A moving target detection method combining MOG2 algorithm and frequency tuned significant region detection(FT)algorithm is proposed.Taking the remote control vehicle walking video captured by the camera as the analysis material,the moving target is detected by MOG2 algorithm on the down sampled image.After binarization and filtering denoising,the region of interest of the moving target is framed,the region of interest is detected by FT algorithm,and the moving target detection is completed by using confidence combined with the two detection results.When the confidence isα=0.6,β=0.4,the average intersection to union ratio of 300 frames is increased from 0.72 to 0.78,which shows that the algorithm can eliminate the influence of background shadow and ghost,so as to effectively improve the detection accuracy of moving targets.

关 键 词：移动目标检测 MOG2算法 FT算法 高斯混合模型 背景减除法 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]