摘要
针对环境迁移、目标被遮挡或姿态变化较大时传统粒子滤波算法的鲁棒性不强的问题,提出一种改进的粒子滤波目标跟踪算法。建立目标模型时,将目标的HSV颜色特征和Uniform LBP纹理特征进行加权融合;粒子重采样过程中,采用加权随机采样方法,将粒子权值作为重采样的影响因子而非决定因子,以提升粒子多样性,降低粒子衰退对目标跟踪的影响;目标被干扰时,采用卡尔曼滤波对目标位置进行偏移校正,以获取目标正确位置;最后采用模板更新策略对目标模板进行实时更新。实验结果表明:相较于传统粒子滤波算法和CMT算法,本文算法对复杂环境中目标被遮挡和姿态变化的情况下都具有较好的鲁棒性。
Given that the robustness of traditional particle filter algorithm for target tracking was not very good, especially in environment migration, occlusion and pose variation, an improved particle filter target tracking algorithm was proposed. When establishing the target model, the target’s HSV color feature and Uniform LBP texture feature were weighted and fused. In the process of particle resampling, the weighted random sampling method was adopted, considering the particle’ s weight as the impact factor of the resampling rather than determinant in order to magnificently improve the diversity of particles and reduce the adverse effects of particles decay. In the case where the target was disturbed, the Kalman filter was used to offset the target position to obtain the correct position of the target. Finally, the introduction of template updating strategy was combined to update target template. The experimental results showed that compared with the traditional particle filter algorithm and CMT algorithm, the proposed algorithm was robust to occlusion and pose variation in complex environments.
作者
高海
韩洋
GAO Hai;HAN Yang(School of Information Science and Techndogy,Zhejiang Sci-Tech University,Hangzhou 310018,China)
出处
《包装学报》
2018年第5期57-64,共8页
Packaging Journal
关键词
粒子滤波
目标跟踪
特征融合
卡尔曼滤波
模板更新
particle filter
target tracking
feature fusion
Kalman filter
template updating
