中文摘要：

针对传统粒子滤波（PF）算法采用单一颜色特征建模跟踪目标性能差的缺陷,提出一种颜色特征与纹理特征相融合的PF目标跟踪新算法。首先,采用一种具有抗噪声和保护纹理边缘的全局中值二值模式（GMBP）纹理算子,对模板图像进行局部差绝对值处理,得到幅值序列模板,将幅值序列模板内的中值作为模板的阈值,与模板邻域比较获得新的纹理图像;然后,与具有光照不变特性的局部二值模式（LBP）纹理算子结合,形成一种（GMLBP）新的纹理描述算子。最后,分别计算GMLBP纹理特征粒子权值和HSV颜色特征粒子权值,并依据权值大小确定融合系数,对纹理特征粒子权值和颜色特征粒子权值进行线性融合,再对融合后粒子权值进行归一化处理,从而得到目标位置状态的最终估计值。对比实验结果表明,相对于单一颜色特征的目标跟踪算法,所提算法捕捉目标位置准确且具有更低的平均跟踪误差,其平均误差降低了近2倍。

英文摘要：

For improving the performance of the traditional particle filter object tracking algorithm based on single color feature,a new particle filter object tracking algorithm based on color and texture feature fusion is presented in this paper.At first,a new global median local binary pattern（GMLBP）texture operator is presented,which is good for noise suppression and texture edges protection.The operator gets an amplitude sequence template by using local differential absolute value for the template image,then center pixels are replaced by median values of the amplitude sequence template,and the new texture image is obtained by being compared with the template neighborhood.Secondly,combining with illumination invariance local binary pattern（LBP）,another new global median local binary pattern（GMLBP）texture description operator is proposed.The fused coefficient is determined according to the different particle weights based on the GMLBP texture characteristic and HSV color characteristics,respectively.The final estimation of target position is proposed via normalizing the fused particle weights.Compared with target tracking algorithm based on single color feature,experimental results show that the proposed algorithm captures the target location accurately and has a lower average tracking error,and the average error is reduced by nearly 2times.