基于粒子流滤波的视觉定位算法

Visual localization algorithm based on particle flow filtering

针对基于粒子滤波的视觉定位算法存在的粒子需求量大以及重采样中粒子贫化而导致的计算量大、定位精度差等问题,提出了一种采用粒子流滤波进行优化的视觉定位算法。该算法通过一个微分方程进行粒子的移动,利用粒子流完成随机样本集从先验分布到后验分布的更新,进而避免了“粒子退化”现象,最后利用参量近似法来求解粒子流滤波,实验结果证明,该算法减少了算法计算量和运行时间,提高了系统的精度。

To solve the problems of large number of required particles,as well as large computation amount and poor positioning accuracy caused by particle dilution in resampling of visual positioning algorithm based on particle filter,a visual positioning algorithm optimized by particle flow filter is proposed.The algorithm uses a differential equation to move the particles,and the particle flow is used to update the random sample set from a prior distribution to a posterior distribution,thus avoiding the“particle degradation”phenome⁃non.Finally,the parameter approximation method is used to solve the particle flow filter.The experiment results show that this algorithm reduces the computational load and running time of the algorithm,improving the accuracy of the system.