基于粒子滤波和自适应模型的目标跟踪算法

为了提高粒子滤波的性能,使用集合卡尔曼滤波对建议分布进行改进,同时提出了用于视频跟踪的自适应融合模型．使用集合卡尔曼滤波结合当前的观测信息构造建议分布,结合当前观测信息对每一个粒子进行集合分析,得到新的建议分布,依据新的建议分布对粒子进行采样,同时在跟踪过程中将颜色特征模型和形状特征模型进行融合,并进行自适应更新．实验结果证明：相对于传统粒子滤波和扩展卡尔曼粒子滤波,使用新的建议分布可以更有效地降低均方根误差,同时自适应融合模型的稳定性要高于使用单一颜色模型．使用新的建议分布和融合模型,可以有效提高粒子滤波的准确性和稳定性．

An object tracking algorithm based on particle filter and adaptive model

To improve the performance of particle filter,ensemble Kalman filter is proposed to construct proposal distribution.And an adaptive fusion model is applied for object tracking.Using ensemble Kalman analysis to build the posterior probability distribution by integrating latest observation information.New particles are resampled from the new proposal distribution.In the tracking process,color model and shape model are fused and updated adaptively.Experimental results show the new proposal distribution can reduce the root mean square error more effectively than traditional particle filter and extended Kalman particle filter.The adaptive fusion model is more stable than single color model.The new proposal and adaptive fusion model can enhance the estimation accuracy and improve the stability of the object tracking.