基于多重单应性变换的改进ViBe算法

Improved ViBe Algorithm Based on Multiple Homographic Transformation

针对目前运动目标检测领域内较流行的ViBe算法不适用于动态相机的问题,根据平面间变换为单应性变换的事实,提出了一种基于多重单应性变换的改进ViBe算法。利用ORB(oriented fast and rotated brief)特征点匹配与随机抽样一致(random sample consensus,RANSAC)算法,计算得到相邻帧的不同平面的变换关系;采用全局光流补偿算法消除运动目标上的特征匹配点对,避免多重单应性变换中包含运动平面的变换;根据多重单应性变换改进ViBe算法的前景判定与背景模型更新,使改进算法适用于运动相机采集的数据。在公开视频数据集CDnet 2014与Complex Background Dataset上进行了相关实验,实验结果证明,改进算法能在尽量保留真实运动目标区域的同时,大幅度消除相机运动带来的影响,且在精度、特异性以及虚警率上表现出色,实时处理效率能达到20帧/s。

Aiming at the problem that ViBe algorithm,which is popular in the field of moving target detection,is not suitable for dynamic cameras,we propose an improved ViBe algo⁃rithm based on multiple homographic transformation accord⁃ing to the fact that the inter⁃plane transform is a homographic transformation.The oriented fast and rotated brief(ORB)feature point matching and random sample consensus(RANSAC)algorithm are used to calculate the transforma⁃tion relationship of different planes of adjacent frames.The global optical flow compensation algorithm is used to elimi⁃nate the feature matching point pairs on the moving target,and avoid the multiple homographic transformation including the motion plane transformation.We improve the foreground decision and background model update of ViBe algorithm ac⁃cording to multiple homographic transformation,so that the improved algorithm is suitable for data collected by motion camera.Experiments are carried out on the public video datas⁃et CDnet 2014 and Complex Background Dataset.The exper⁃imental results show that the improved algorithm can greatly eliminate the impact brought by camera motion while preserv⁃ing the real moving target region,and it has excellent perfor⁃mance in accuracy,specificity and false alarm rate.The algo⁃rithm can achieve a real⁃time processing efficiency of 20 fps.