摘要： 基于多目标跟踪中的遮挡问题与互动社会模型间的联系,提出利用具有群间相互动态信息的多群社会模型改进简化粒子群优化算法,并用于多目标跟踪。在粒子群的多样性基础上初始化新群,预测目标速度。结合多群的优化,在连续性信息和目标之间修改简化粒子群优化的更新方程式,即更新目标速度和目标位置。为了适应目标进入和离开现场,构建初始化新群和终止迭代的策略。CAVIAR数据集、PETS2009数据集和Oxford数据集上的实验结果表明,相比于颜色粒子滤波算法、基于直方图的算法、局部稀疏法和高斯密度函数法,提出的算法在多目标跟踪精度方面至少提高了10%,大多数跟踪轨迹的数量增加了约8%,能鲁棒跟踪多个目标。
关键词:
多目标跟踪,
简化粒子群优化,
社会模型,
终止策略,
鲁棒
Abstract: Based on the connection of occlusion problem in multi-target tracking and social interaction model,an improved Simplified particle Swarm Optimization(SSO) algorithm of multi-group social model with inter-group dynamic information is proposed,which is used for multi-target tracking.On the basis of diversity of particle swarm,a new swarm is initialized,predicting speed of target.Combined with multi-swarm optimization,the update equation of SSO is modified by continuity information and targets,that is the target speed and position are updated.In order to adapt the target’s entering and leaving the scene,strategies for initializing new swarms and terminating iteration are constructed.The effectiveness of the proposed algorithm is verified by experimental results on CAVIAR data set,PETS2009 data set and Oxford data set.Compared with color-based particle filtering algorithm,histogram-based algorithm,local sparse algorithm and Gaussian density function algorithm,the proposed algorithm improves at least 10% in multi-object tracking accuracy,and the number of most tracking trajectory increases by about 8%.The proposed algorithm can track multiple targets robustly.
Key words:
multi-target tracking,
Simplified particle Swarm Optimization(SSO),
social model,
termination strategy,
robust
中图分类号: