基于XCS-LBP纹理背景建模算法的运动目标检测

Background modeling algorithm based on XCS-LBP texture feature

传统纹理特征的背景建模算法具有模型的维数过高、计算复杂等缺点,加大了运动目标检测的难度,提取精确细致的纹理特征建立背景模型,并且提高算法的处理速度,是运动目标检测技术的关键。文章对比了几种经典的纹理算子,采用了提取纹理信息更加细致的纹理提取算子,使用统计学习方式建立背景模型,又利用像素的空间信息,增加了随机更新机制去更新背景模型,提出了基于扩展中心对称局部二值纹理模式(extended center-symmetric local binary pattern,XCS-LBP)纹理特征背景建模算法并通过实验验证该算法检测效果好,避免了模型的维数过高,减少了复杂计算,处理速度满足实时处理的需求。

Traditional background modeling algorithms based on texture features have the disadvantage of high dimension and complex calculation, which increase more challenges to moving object detection. Extracting precise and detailed textural feature used for building background model, and improving processing speed of the algorithm are two key factors of moving target detection technology. This paper compares some classical texture operators and adopts texture extraction operator which can extract more detailed texture information, then builds background model using statistical learning method. The random update mechanism is added to update background model using space information of pixels. Then this paper proposes a background modeling algorithm based on XCS-LBP （ Extended Center-Symmetric local Binary Pattern, XCS-LBP） texture feature. Experimental results show that the algorithm proposed in this paper works well, avoids high dimension and complex calculation, and has fast processing speed which meets requirement of real-time processing.