采取阶段性改进的全新ViBe目标检测算法

针对ViBe (Visual Background extractor)算法在目标检测过程中易产生鬼影问题和检测目标不完整问题,从ViBe算法处理过程的主要阶段出发,提出一种全新的ViBe目标检测算法.首先,在模型初始化阶段,利用前m帧视频序列对应像素点的均值构建背景模型,同时将原算法的8邻域改为24邻域进行样本选取以及动态调整匹配半径;然后,在目标检测阶段,引入最大类间方差法来计算当前图像帧的最佳分割阈值,进而对前景像素进行二次判别;其次,在背景模型更新阶段,根据背景变化快慢程度动态地调整更新因子;最后,对获得的前景图像进行形态学处理得到最终的前景目标.实验结果表明,改进后的ViBe算法使鬼影问题得到有效解决,目标检测的准确度和完整度也有大幅提高.

A new target detection algorithm of ViBe based on phased improvement

In view of the problems of ghost and incomplete detection in the process of target detection in the visual background extractor (ViBe) algorithm, this paper proposes a new algorithm of target detection of ViBe based on phased improvement.Firstly, in the initial phase of the model, the background model is constructed using the average value of the corresponding pixels of the first m frames of video sequence.At the same time, the 8 neighborhood of the original algorithm is changed to 24 neighborhood for sample selection and dynamic adjustment of the matching radius; then in the target detection stage, the maximum inter class difference method is introduced to calculate the best segmentation threshold of the current frame, and then the foreground pixels are discriminated twice; Secondly, in the phase of background model updating, the size of updating factor is dynamically adjusted according to the speed of background change; finally, the final foreground target is obtained by morphological processing of the obtained foreground image.Experimental results show that the improved ViBe algorithm not only solves the ghost problem effectively, but also improves the accuracy and integrity of target detection.