基于改进高斯混合模型及纹理的阴影去除算法

摘要

为准确从视频中提取出不含阴影的运动目标,提出一种基于改进高斯混合模型及纹理的阴影去除算法.采用自适应高斯模型个数选取策略及自适应学习率参数取值方法对经典高斯混合模型进行改进,将改进后的高斯混合模型与三帧差法结合提取运动区域,使用基于梯度方向的纹理相关性算法移除运动区域的阴影.实验结果表明,该算法能快速、精准地建立并实时更新背景模型,在保证运动目标实际轮廓完整性的前提下,准确去除光照条件下带来的阴影.%To accurately extract the foreground object without shadow from the video,a shadow removal algorithm based on improved Gaussian mixture model and texture was presented.The self-adaptive number of Gaussian model selection strategy and self-adaptive learning rate parameter value choosing method were adopted to improve the classic Gaussian mixture model,the improved Gaussian mixture model was combined with the three frame difference to extract the moving area,the shadows were discriminated from objects using texture correlation algorithm based on gradient direction.Experimental results indicate that the proposed method can establish background model rapidly and accurately as well as update the background in real-time,meanwhile,on the premise of ensuring contour integrity of moving objects,the method is able to remove the shadow effectively in the sunlight condition.