针对运动目标检测中单特征背景模型存在的局限性,如基于颜色特征的背景模型对光照和阴影敏感、基于纹理特征的背景模型易产生空洞,提出了一种以置信度融合RGB颜色特征和SILTP(scale invariant local ternary pattern)纹理特征的运动目标检测方法.以像素点SILTP纹理信息值和RGB颜色信息值及它们各自的置信度构建背景模型.分别计算当前像素点与背景模型的纹理差异度和颜色差异度,通过置信度融合的方法计算当前像素与背景模型的总体差异度,以达到更好的融合效果.采用ViBe算法的更新策略更新背景模型.通过前景像素点8邻域内的背景像素点的统计方法去噪点.Wallflower和Data 2014数据集上的实验结果表明,所提出的融合方法能有效抑制阴影,对光照有良好的鲁棒性,在复杂动态背景下能取得良好的效果.%Background model based on single feature has limitations.Color feature is sensitive to illumination and shadow and texture feature may lead to cavity.To solve the problem,this paper proposed a method of moving object detection by fusing SILTP(scale invariant local ternary pattern)texture feature and color feature based on confidence.It built the background model with SILTP、RGB value and their confidences.Calculated the texture and the color difference degree between current pixel and its model respectively,then calculated the overall difference with confidences to get better results.Still used the updating stragedy of ViBe.Reduced error points by counting the number of pixels that was the background among the eightneighborhood of each foreground pixel.The results of test on wallflower and Data 2014 dataset shows that the proposed algorithm can inhibit shadow effectively and have good robustness to the change of illumination,also it performs good in complex dynamic background.
展开▼
