一种改进视觉背景提取（ViBe）算法的车辆检测方法

视觉背景提取(visual background extractor，ViBe)算法应用在车辆检测时存在一个比较明显的缺点，即当视频第1帧中存在待检测的移动车辆时，在后续帧的车辆检测过程中，对应第1帧中车辆的位置处会出现鬼影并且鬼影会持续一段时间才会彻底消失，从而干扰后续帧的检测效果。提出一种改进的ViBe建模方法，新方法在前n帧中实现初始模型的初始化，并结合ViBe算法的更新方法进行模型更新。在不同分辨率、不同场景的视频中对原算法和提出的改进方法进行对比实验，实验结果表明，在第1帧中不包含车辆和包含车辆2种情况下，提出的改进的算法都能有效地检测出移动车辆且不会产生鬼影的问题。因此，改进方法比原算法更有效和实用。

A vehicle detection method based on improved ViBe algorithm

There is a flaw when ViBe algorithm is used to detect vehicles, that is, the ghost of vehicles would exist over a period in vehicle detection if moving vehicles exist in the first frame and leave in the following frames. Aiming at ghost removing, an improved ViBe modeling method is proposed in this paper. The initialization model would be initialized in the first n frames and updated according to original ViBe update algorithm in the proposed method. The comparative experiments between the original ViBe algorithm and the proposed methods are taken on several videos with different scenes and resolutions. Experiment results show the proposed method can detect moving vehicles as well as remove the ghosts effectively although there are moving vehicles in the first frame of the video. Therefore, the proposed method would be more effective and practical than the original ViBe algorithm.