摘要
针对传统ViBe算法在复杂背景下检测运动目标时会出现鬼影、阴影、误检等问题,提出了一种改进的ViBe运动目标检测算法,称为GS-ViBe算法。在GS-ViBe背景模型初始化阶段,利用最大后验估计法确定每个像素点的最佳高斯分布数目,使其形成多帧融合背景来代替ViBe的单帧背景初始化方法,从而消除鬼影;在GS-ViBe前景检测阶段,增加多特征融合阴影检测过程,并将其检测结果和ViBe前景目标融合,得到消除阴影后的前景目标;最后,在GS-ViBe背景模型更新阶段,引入动态更新因子代替固定更新因子,使得背景可以自适应更新,从而降低目标的误检率。在多种复杂背景下与传统ViBe算法对比发现,GS-ViBe算法召回率提高了37.74%,准确率平均提高了19.83%,误检率平均降低了52.57%,表明GS-ViBe算法可以有效消除鬼影、阴影、误检的干扰,获取到完整的前景目标。
Aiming at the problems of ghosts,shadows,and false detections occur when traditional ViBe algorithm detects moving targets in complex backgrounds,an improved ViBe moving target detection algorithm was proposed,which was called GS-ViBe algorithm.In the initialization stage of GS-ViBe background model,the maximum posteriori estimation method was used to determine the optimal number of Gaussian distributions of each pixel to form a multi-frame fusion background instead of single-frame background initialization method of ViBe,so that the ghosts were eliminated.In the GS-ViBe foreground detection stage,the multi-feature fusion shadow detection process was added,and the detection results were fused with ViBe foreground targets to obtain the foreground targets after eliminating shadows.Finally,in the GS-ViBe background model update stage,a dynamic update factor was introduced instead of a fixed update factor,so that the background could be updated adaptively,thereby reducing the false detection rate of targets.In comparison with traditional ViBe algorithm in a variety of complex backgrounds,it is found that the recall rate of GS-ViBe algorithm is increased by 37.74%on average,the accuracy rate is increased by 19.83%on average and the false detection rate is reduced by 52.57%on average.It shows that the GS-ViBe algorithm can effectively eliminate the interference from ghosts,shadows and false detections,which obtains the complete foreground targets.
作者
贾澎涛
侯长民
李娜
JIA Pengtao;HOU Changmin;LI Na(College of Computer Science and Technology,Xi'an University of Science and Technology,Xi'an 710054,China)
出处
《应用光学》
CAS
北大核心
2023年第5期1045-1053,共9页
Journal of Applied Optics
基金
国家自然科学基金(62002285)。
关键词
运动目标检测
阴影检测
ViBe
最大后验估计
多特征融合
moving target detection
shadow detection
ViBe algorithm
maximum posteriori estimation
multi-feature fusion
作者简介
贾澎涛(1977-),女,博士,教授,硕士生导师,主要从事人工智能、机器学习、目标检测研究。E-mail:jiapengtao@xust.edu.cn;通信作者:侯长民(1997-),男,硕士研究生,主要从事目标检测、图像处理、深度学习研究。E-mail:houchangmin333@163.com。
