针对ViBe算法易受光照以及动态背景影响，在运动目标检测过程中容易出现鬼影、阴影和运动目标空洞问题，提出一种改进ViBe和多特征融合的运动目标检测算法。首先，初始化背景建模采用鲁棒主成分分析(RPCA)方法，解决首帧产生鬼影问题；其次，引入标准离差率和帧间均速测量值能够自适应改变匹配半径和更新速率，从而适应动态背景变化情况，消除空洞现象；最后，融合HSV、LBP和Gabor特征对阴影进行检测并去除。在公开的CDnet 2014数据集进行实验，结果表明，本文算法能够适应动态背景变化，有效解决了鬼影和阴影问题，在多种场景下能完整提取目标。
For the ViBe algorithm is susceptible to lighting and dynamic background, ghosts, shadows and holes in the moving target detection process, an improved ViBe and multi-feature fusion motion target detection algorithm is proposed. Firstly, robust principal component analysis (RPCA) is used to initialize background modeling to solve the ghosting problem in the first frame. Secondly, the standard deviation rate and interframe average speed measurements are introduced to adaptively change the matching radius and update rate, so as to adapt to dynamic background changes and eliminate holes. Finally, HSV, LBP, and Gabor features are combined to detect and remove shadows. Experiments on an open CDnet 2014 dataset show that the algorithm can adapt to dynamic back-ground changes, effectively solve ghost and shadow problems, and extract targets completely in multiple scenarios.
