融合ViBe改进算法和图像均值的运动目标检测

Fused ViBe improvement algorithm and image averaging for motion target detection

摘要:

针对视频前景提取(ViBe)算法在模型初始化时因前景像素干扰导致的“鬼影”问题，面对复杂背景环境的更新策略问题，提出利用图像像素均值作为参考对ViBe算法模型进行初始化优化；同时，提出随背景模型复杂度变化的自适应更新策略。利用邻域像素和连续帧背景像素的相似性进行背景模型初始化；然后通过计算样本间各像素的方差判定背景模型是否稳定，建立自适应的更新策略；最后提取运动目标。通过CDnet2014数据集验证表明：该算法有效改善了“鬼影”现象，提高了背景模型在复杂环境下的鲁棒性，各项客观评价指标也有所提升。

Abstract:

In view of the "ghost" problem caused by the interference of foreground pixels in the model initialization of Visual Background Extractor(ViBe) algorithm, and facing the problem of updating strategies for complex contextual environments, the initial optimization of the ViBe algorithm model using the image pixel mean as a reference is proposed. At the same time, an adaptive updating strategy varied with the complexity of the background model is proposed. Firstly, the background model is initialized by using the similarity between neighborhood pixels and continuous frame background pixels. Then whether the background model is stable is determined by calculating the variance of each pixel between the samples, in order to establish the adaptive update strategy. Lastly, the moving target is extracted. The validation of CDnet2014 data set shows that this algorithm can effectively improve the phenomenon of "ghost". It not only improves the robustness of the background model in the complex environment, but also improves the objective evaluation indexes.