Multi-feature Fusion Tracking Based on A New Particle Filter
Multi-feature Fusion Tracking Based on A New Particle Filter
复制标题
DOI：
10.4304/jcp.7.12.2939-2947
复制
发表时间：
2012-01
期刊：
影响因子：
--
通讯作者：
Wei Li-;Jie Cao;Di Wu
中科院分区：
文献类型：
--
作者：
Wei Li-;Jie Cao;Di Wu
提出了一种新的粒子滤波算法用于非线性系统的状态估计。该算法利用积分剪枝因子对积分点进行优化和重组，从而实现了基于正交卡尔曼滤波的算法。新算法利用剪枝正交卡尔曼滤波产生最优的建议分布函数，较好地克服了粒子退化现象。在改进的粒子滤波框架中，使用颜色和运动边缘特征作为观察模型。通过D-S证据理论融合特征权重，有效避免了单一颜色特征在光照突变、姿态变化和相似特征遮挡时产生的鲁棒性差的问题。实验结果表明，该方法对目标的跟踪具有较强的鲁棒性，在复杂场景下具有较好的性能。
A new kind of particle filter is proposed for the state estimation of nonlinear system. The proposed algorithm based on Quadrature Kalman Filter by using integral pruning factor, which optimizes and reorganizes the integration point. New algorithm overcomes the particle degeneration phenomenon well by using Pruning Quadrature Kalman Filter to produce optimized proposal distribution function. In the improving particle filter framework, using color and motion edge character as observation model. Fusing feature weights through the D-S evidence theory, and effectively avoid the questions of bad robust produced by the single color feature in the illumination of mutation, posture change and similar feature occlusion. Experiment results indicate that the proposed method is more robust to track object and has good performance in complex scene.