与视频背景颜色相似的纹理缺乏前景目标检测方法

FOREGROUND OBJECT DETECTION METHOD WITH BACKGROUND COLOR SIMILARITY AND POOR TEXTURE

现有的视频运动目标检测方法无法有效跟踪与背景颜色相似且纹理缺乏的运动目标,对此,提出一种基于非下采样小波变换的LBP(UW-LBP)纹理特征提取新方法。对当前图像和背景图像进行三层非下采样小波变换;对每个小波变换子图提取LBP纹理特征,为了提高运算速度,没有采用LBP直方图,而是用一个8×n位的二进制向量定义UW-LBP描述子,并用海明距离度量局部纹理的差异;提出一个从像素级到图像块级的层次的运动目标检测策略。实验结果表明,所提出的算法能够有效地检测与背景颜色相似的纹理缺乏运动目标,并对噪声和环境变化有良好的鲁棒性。

Aiming at the existing problem of the segmentation of foreground objects with camouflaged color and poor texture in video surveillance, a novel method is proposed to detect foreground object with camouflaged color and poor texture. This paper proposed a local binary pattern descriptor based on un-decimated wavelet （UW-LBP）. Three layers un-decimated wavelet transform was used for image and background images. To reduce the consumption of computation time, the UW-LBP descriptor was represented by an 8×n bit binary vector, instead of LBP histogram. Hamming distance was used for measuring the difference of local texture. A hierarchical foreground detection scheme of the pixel level to block level is proposed. Experimental results show that the proposed approach can effectively detect moving objects with camouflaged color and poor texture and improve the robustness against environmental changes and noise.