融合多尺度变换的改进Vibe算法

Improved Vibe Algorithm Integrated with Multiscale Transformation

针对视觉背景提取(Vibe)算法在目标检测过程中存在鬼影且易受动态背景干扰等问题,提出了一种基于多尺度空间的Vibe算法。在建立背景模型前,对输入的视频序列进行金字塔变换,得到顶层、中层、底层3种不同分辨率的图像;再在不同分辨率下进行Vibe前景检测,并对检测结果进行融合,减少了动态背景的影响,同时提出了一种鬼影消除策略,结合帧间信息,加入二次判断策略,加快了鬼影的消除;最后,为了更好地适应动态环境,提出了一种背景复杂度量,根据背景的复杂程度,自适应地调整阈值。实验结果表明:经改进的算法加快了鬼影的消除,对动态环境噪声的干扰有良好的稳健性。

A visual background extracting （Vibe） algorithm based on multiscale space is proposed to solve the problem that there exist ghosts and dynamic background disturbance in the target detection process for the conventional Vibe algorithms. Before the background model is established, the input video sequence is pyramid transformed to obtain three different resolution images at the top, middle, and bottom layers. Then the Vibe foreground detection is performed at different resolutions, and the detection results are {used to reduce the influence of dynamic background. At the same time, a ghost elimination strategy is proposed which enhances the ghost elimination by using the information between frames and by adding a second judgement. Finally, in order to well adapt to the dynamic environment, a metrology under a complex background is proposed, in which the threshold can be adaptively adjusted according to the complexity of backgrounds. The experimental results show that the improved algorithm accelerates ghost elimination and has strong robustness to dynamic environmental noise disturbances.