基于最大后验概率密度的粒子过滤器跟踪算法

Kalman滤波的弱点是它无法解决非线性、非高斯问题的跟踪。为此提出了一种新型的跟踪算法,粒子过滤器算法。该算法采用加权的粒子集模型表示状态的分布,迭代跟踪状态的变化。其优点是它可以适应复杂环境的重叠和遮挡情况,且能同时跟踪多目标。采用最大后验概率模型确保了状态判断和估计的准确性。对重采样的分析减少了算法对噪声的敏感。并把样本安排在目标可能出现的区域。在眼睛跟踪系统上实现了该算法。仿真结果表明MAP模型在精度上与传统的方法比较提高7%。眼睛跟踪的结果证实了仿真的结果。

Particle filter tracking algorithm based on maximum a posteriori

The weakness of Kalman filter is that it is inadequate to solve the problem of non-linear and non-Gaussian model. A new model with importance weights, density assisted particle filter algorithm is proposed. This method represents the distribution of the states by weighted samples and iterates the change of states. The virtue of this algorithm is that it can adapt complex interactions with overlap and ambiguities and track to multiple objects simultaneously. Maximum A Posteriori (MAP) mode is exploited the statistical dependence between objects to provide accurate decisions and state estimation. Analysis of resample techniques reduces the sensitivity of the algorithm to noises. Weight correspondence to sample should be arranged in a distribution around the most possible area of object. The algorithm is implemented on the system of eye tracking. Simulation results suggest that MAP mode can potentially provide 7% improvement in accuracy over conventional mode. Results of eye tracking support the simulation studies.