基于双特征的多模块目标跟踪算法

针对运动目标在受到严重遮挡时难以被精确跟踪的问题,提出一种融合颜色和LBP (local binary pattern)纹理特征的多模块跟踪算法.综合考虑目标与背景的特征显著性和相似性两个因素建立比值关系进行量化分析,选取了能够最大程度区分前景目标和背景的颜色空间特征,并结合LBP纹理特征建立概率分布直方图.利用卡尔曼滤波器预测均值漂移算法的初始迭代位置.引入相似度因子来定义新的遮挡判决准则,自适应采用多模块模型进行跟踪.仿真实验结果表明了该算法的有效性.

Multi-block target tracking algorithm based on dual feature

To solve the problem of moving target tracking under severe occlusion,a novel multi-block tracking algorithm is proposed by fusing the color and LBP(local binary pattern) texture feature.Target and background feature conspicuousness and similarity are used to set up ratio relation for selecting the color feature that could furthest differentiate the prospect target and background,and probability histogram is set up by color and LBP texture feature.The Kalman filter is adopted to predict the initial position in mean shift algorithm.The similarity factor is proposed for defining the occlusion criteria to adopt multi-block model adaptively.The effectiveness of the proposed algorithm is demonstrated by simulation experiments.