一种基于多特征融合的目标跟踪新方法  

A Novel Target Tracking Method Based on Multi-Feature Fusion

作　　者：郑宇杰[1] 孙勇成[1] 

机构地区：[1]中国电子科技集团公司第28研究所,南京210007

出　　处：《信息化研究》2015年第3期14-17 62,62,共5页INFORMATIZATION RESEARCH

摘　　要：仅利用单一的目标特征进行跟踪是大多数跟踪算法鲁棒性不高的重要原因。文章提出了一种基于粒子滤波框架的多特征融合跟踪方法。该方法以log-Gabor滤波器作为粒子判别级,同时融合局部二元模式(Local Binary Pattern,LBP)特征和方向梯度直方图(Histograms of Oriented Gradients,HOG)特征在内的粒子加权级,实现目标跟踪。运用log-Gabor滤波器对粒子做出总体评估,同时结合LBP和HOG局部特征对目标做出细节把握,有效突出后验概率分布中的峰值状态。该算法在一种特征受到背景干扰导致目标鉴别能力降低时,其他特征仍能稳定可靠的跟踪目标。实验结果表明,该算法能快速去除无效粒子,实现复杂背景下的目标跟踪。Only single target feature used in target tracking is the important reason of many algorithms without robust performance.In order to increase the effectiveness of target tracking,a novel algorithm based on particle filter is proposed.Invalid particles are removed in discrimination stage by log-Gabor filter.Then,Local Binary Pattern(LBP)and Histograms of Oriented Gradients(HOG)features are used in target tracking.Log-Gabor filters have good property on frequency response,such that decides the effectiveness of particles.In additional,LBP and HOG features are combined to demonstrate the local features.Experiment's results demonstrate performance of the proposed and realize robust target tracking effectively.

关 键 词：目标跟踪 粒子滤波 LOG-GABOR滤波 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]