改进的多特征粒子滤波目标跟踪算法研究

Research on Improved Multi-feature Particle Filter Target Tracking Algorithm

摘要:

针对在复杂环境下多特征融合的粒子滤波算法跟踪精确度低的问题，提出一种改进的多特征融合算法；该算法采用二阶中心差分卡尔曼滤波方法来实现建议分布函数的优化，在重要性采样中融入最新的测量信息，提高了粒子的使用效率，并引入动态模板更新机制对目标模板实时更新；在多特征融合策略上利用基于粒子滤波框架下的EM算法适用于不同数量样本集的特点求解状态估计，不仅避免因计算特征权重产生误差，而且提高了算法的实时性；滤波器仿真实验结果表明，在一维非线性模型下对比其它改进粒子滤波算法，本文提出的方法性能最优；在基于视频序列的目标跟踪实验中，通过比较本文算法在不同特征、不同采样粒子数量条件下的性能对比验证本文算法的有效性；最后通过一系列不同环境下的跟踪实验证明，本文算法对复杂条件下的目标跟踪具有较高的精度和鲁棒性。

Abstract:

In order to solve the problem that the particle filter(PF) algorithm based on multi-feature fusion under complex environment did not offer a high accuracy of tracking technique, an improved multi-feature fusion algorithm was proposed. The research proposed the second-order central difference Kalman filter(SO-CDKF) to generate the proposal distribution function which can match the true posterior distribution more closely. The latest observation information was fused into the importance sampling to improve the efficiency of particles. Meanwhile, the introduction of template updating strategy was combined to update target template in real time. In multi-feature fusion strategy, Expectation-Maximization(EM) algorithm based on PF framework was used to get the state estimation of different quantity of particle sets, so as to avoid errors caused by calculating the weights of multi-feature, and improve the real-time performance. Filter simulation results show that the proposed method has the best performance compared with the other improved PF algorithms in one-dimensional nonlinear model. In the experiment of target tracking based on video sequence, the effec-tiveness of the proposed algorithm is verified by comparing it’s performance of different features and quantity of particles. Finally, a series of tracking experiments in different environments show that the pro-posed algorithm has high accuracy and robustness for target tracking under complex conditions．