基于试探采样的自回馈目标跟踪算法

针对粒子传播过程中因欠缺观测信息而导致退化现象和异常粒子，文中提出一种基于试探采样的自反馈目标跟踪算法。该算法在当前帧完成采样后向前试探采样粒子，并且反馈到当前帧，此举是利用未来帧提前采样形式把观测信息融入到状态转移模型中，从而使概率密度分布逼近真实值。分析上下帧间粒子权值关系，舍弃异常元素，进行不完全重采样，在缓解退化问题同时保持样本集多样性。目标状态估计采用加权-最大后验准则，提高了目标跟踪精确度与稳定性。实验结果表明所提算法提高了状态空间质量，相比其他算法具有更好的跟踪性能。

Self-feedback Object Tracking Algorithm Based on Tentative Sampling

In order to solve the problems of degeneracy phenomenon and abnormal particles due to the lack of observation information in the particles diffusion process, This paper proposed a novel self-feedback object tracking algorithm based on tentative sampling. The proposed method drew forward a set of particles tentatively after the sampling of the current frame and fed back to it. This step can introduce the observation information into the state transition equation by sampling in the future frame, so the probability density distribution can approximate the true point well. In the analysis of the weight relationship between the consecutive frames, this algorithm removed the abnormal factors and carried out part resampling the partial particles. This method can not only alleviate particle degeneracy, but also improve the diversity of particles. WS-MAP criterion introduced can improve the accuracy and stability of object tracking in state estimation. Experimental results show that the proposed method can improve the quality of the state-space, and has better tracking performance than the other tracking algorithms.