一种基于纹理和颜色置信融合的运动目标检测方法

发布时间：2018-05-12 08:23

本文选题：运动目标检测 + 颜色特征　；参考：《计算机应用研究》2017年07期

【摘要】：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Vi Be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。
[Abstract]:The background model based on color features is sensitive to illumination and shadow, and the background model based on texture features is prone to generate holes. A method of moving target detection based on confidence fusion of RGB color feature and SILTP(scale invariant local ternary pattern) texture feature is proposed. The background model is constructed by pixel SILTP texture information, RGB color information and their confidence. The texture and color differences between the current pixel and the background model are calculated respectively. The overall difference between the current pixel and the background model is calculated by the method of confidence fusion to achieve a better fusion effect. The update strategy of Vi be algorithm is used to update the background model. The experimental results on background pixels in foreground pixel 8 neighborhood and Data 2014 dataset show that the proposed fusion method can effectively suppress shadows and is robust to illumination. Good results can be obtained under the complex dynamic background.
【作者单位】：武汉理工大学计算机科学与技术学院;
【基金】：国家自然科学基金资助项目(61003130) 国家科技支撑计划资助项目(2012BAH33F03) 湖北省自然科学基金资助项目(2015CFB525) 武汉市科技计划创新团队项目(201307020402005)
【分类号】：TP391.41

【相似文献】