复杂背景下改进视觉背景提取的前景检测算法

摘要

针对视觉背景提取模型存在的鬼影抑制效果差、动态背景适应能力不足等问题,提出了一种改进的视觉背景提取模型算法.在模型构建阶段,该算法充分融合时空域信息初始化背景模型,避免了样本的重复选取,提高了鬼影抑制能力;在像素分类阶段,根据背景动态程度,引入自适应距离阈值代替全局固定阈值,增强了模型对动态背景的适应性;在背景更新阶段,对连续多帧判定为前景的像素点进行阈值判断,并及时更新到背景模型,消除了运动背景与静止前景造成的虚影现象.多个公开视频数据的测试结果表明,该算法相比典型算法在复杂背景下检测的准确性和鲁棒性都有了很大提高.%Aiming at the poor effect to suppress the ghost and the low ability to adapt to the dynamic background of visual background extractor model,this paper proposed an improved algorithm based on visual background extractor.In the modeling stage,this algorithm fully fused the temporal and spatial information to initialize the background model,which avoided the repetitive selection of the samples and improved the ability to suppress the ghost.In the pixel classification stage,according to the change degree of dynamic background,it used the self-adaptive distance threshold instead of the global fixed threshold to improve the ability to adapt to the dynamic background.In the background updating stage,in order to eliminate the ghost caused by moving background and still foreground,it updated the pixels which were classified as the foreground in the continuous frames as the background after threshold judgment.The test results of multiple open video data show that the proposed algorithm greatly improve the accuracy and robustness of the detection in the complex background compared with other typical algorithms.