基于自适应的ViBe运动目标检测方法

Moving Object Detection Methods Based on Adaptive ViBe

针对视觉背景提取(VIsual Background Extractor,ViBe)运动目标检测算法参数固定且无法消除鬼影干扰的问题,提出了一种自适应策略的ViBe运动目标检测算法。在样本选取上,扩大样本邻域且随机生成样本,避免了邻域的重复选取导致的错误分类。对于ViBe匹配半径和背景更新因子的固定参数设置,自适应地根据背景的动态程度设置匹配半径和利用运动速度动态调整更新因子。为了消除鬼影,通过迭代法获得二次判别的最优阈值,过滤掉误判区域,可以快速将鬼影区域重新判别为背景。在2个公共视频数据集上,与具有代表性的6种算法进行比较,实验表明所提出方法的有效性,且对复杂场景下的运动目标检测具有鲁棒性。

In order to solve the problems of fixed parameter setting and ghost,an adaptive strategy for the ViBe(Visual Background Extractor)motion object detection algorithm is proposed,which includes three improved adaptive methods.First,for the sample selection,it expands the sample neighborhood with the uniform random number,which can avoid to select pixels repeatedly and reduce pixel error classification.Then for fixed match radius and background updator in ViBe,adaptive strategies are used respectively that matching radius are set by the degree of dynamic background and background updator is dynamically regulated by the motion velocity.Finally the optimal threshold of quadratic discriminant obtained by an iterative method is used to filter out the misjudgment area and eliminate the ghost areas.Compared with frame difference method,GMM(Gaussian Mixture Model),CodeBook,LSD(Low-rank and Sparse Decomposition),DECOLOR(Detecting Contiguous Outliers in the Low-rank Representation),and ViBe algorithm on the two public video datasets,Change Detection and LASIESTA,the proposed method has a better performance and it is robust for moving object detection in complex background.