多特征融合的退火粒子滤波目标跟踪

针对传统粒子滤波的建议分布没有利用到当前观测信息的缺点，提出了一种基于多特征融合的退火算法来改进建议分布的粒子滤波跟踪方法。该方法解决了高维状态下计算量大和粒子数匮乏问题。采用退火方法在蒙特卡洛重要采样范围内产生更好的建议分布，并用退火似然性抽样来代替简单的先验概率抽样。在似然逼近中，应用颜色和边缘相融合的图像特征属性在不同的退火层加权来产生权重功能函数。用该方法对复杂背景下和存在遮挡情况下的运动目标进行跟踪，结果表明该方法有较高的跟踪精度和较强的稳定性。

Multi-featured integration annealing particle filtering target tracking

Proposal distribution of traditional particle filtering has the shortcomings which is lacking of utilizing current observational information.In order to improve the performance of particle filter for target tracking,a particle filter tracking method based on multi-featured integration annealing algorithm is proposed to improve the proposal distribution.The proposed solu- tion gives an answer to the large amount of calculation and lacking of particles under high-dimensional conditions.With adoption of the approach, the better proposal distribution can be generated within the scope of Monte Carlo importance sampling range, and the simple priori probability sampling can also be replaced by annealing likelihood sampling.In the likelihood approximation, image feature attribute of colors and edges integration is applied to generate weight function at different annealing layer by weighing.Experiment shows the method is of higher tracking accuracy and stronger stability when tracking moving objects is under complex situation and occlusion circumstance.