摘要
采用概率假设密度(PHD)粒子滤波进行多目标跟踪时,各时刻的目标状态表现为大量的加权粒子,需以一定方法从该粒子近似中提取出来。该文提出一种增强的目标状态提取方法,先以k-means算法对粒子进行空间分布的聚类,再于各类中寻找粒子权的峰值位置作为目标状态的估计。仿真结果表明:由于综合利用了粒子的权值和空间分布信息,该算法具有比现有算法更小的目标状态估计误差。
Probability Hypothesis Density(PHD) filter has emerged as one of powerful tools for multi-target tracking.In the Sequential Monte Carlo(SMC) implementation of it,the filter's output is particle approximation of PHD,so some special algorithm is needed to extract the target states from those particles.In this paper,an improved algorithm is proposed.Firstly particles are clustered by their positions using the k-means algorithm,and then the positions with maximum of particles' weight are searched and estimated in each cluster as the targets' positions.Because the information of both particles' weight and spatial distribution are utilized,confirmed by simulation results,the new algorithm can provide estimation of the targets states more accurately.
出处
《电子与信息学报》
EI
CSCD
北大核心
2010年第11期2691-2694,共4页
Journal of Electronics & Information Technology
关键词
多目标跟踪
贝叶斯滤波
粒子滤波
概率假设密度
聚类
Multi-target tracking
Bayes filtering
Particle filter
Probability Hypothesis Density(PHD)
Clustering