CAJ | 학술논문

提出了一种新的融合多特征的基于改进模拟退火粒子滤波跟踪算法。首先，针对重要性采样粒子滤波算法中重要性抽样密度函数未考虑最近观测值，不能有效逼近真实后验密度函数的问题，通过采用改进的模拟退火(SA)方法优化重要抽样密度函数，并利用不同温度下扰动函数和Metropolis准则克服粒子匮乏缺陷；同时，针对SA方法在粒子滤波视觉跟踪应用上效率不高的缺陷，对经典模拟退火算法进行改进，降低了参数选择的敏感性，保持了原算法全局寻优的优点，提高了算法的速度。
제출료일충신적융합다특정적기우개진모의퇴화입자려파근종산법。수선，침대중요성채양입자려파산법중중요성추양밀도함수미고필최근관측치，불능유효핍근진실후험밀도함수적문제，통과채용개진적모의퇴화(SA)방법우화중요추양밀도함수，병이용불동온도하우동함수화Metropolis준칙극복입자궤핍결함；동시，침대SA방법재입자려파시각근종응용상효솔불고적결함，대경전모의퇴화산법진행개진，강저료삼수선택적민감성，보지료원산법전국심우적우점，제고료산법적속도。
This paper presents a novel particle filter tracking algorithm that fuses multiple cues using improved simulated annealing method.The importance probability density function adopted by the algorithm of sequential importance sampling particle iflter does not take into account the updated observation,so that it can not approximate posterior density well.Therefore,an improved simulated annealing method is introduced to better approach the real probability density function.The problem of particle impoverishment can be solved by utilizing disturbing function and by the rule of metropolis. Meanwhile,in order to obviate the defect of low efficiency that SA method performed in particle filter target tracking application,the classical simulated annealing algorithm is improved,reducing the sensitivity of parameter selection and maintaining the advantage of global optimization in the original algorithm.Besides that,the speed of the algorithm proposed in the paper runs faster.