基于目标区域预检测模型的粒子滤波小目标跟踪算法

Particle filter for small target tracking based on target region pre-detect model

针对低信噪比光电图像序列中小目标跟踪计算量大的问题,从降低粒子滤波算法中复杂度出发,提出一种基于目标区域预检测模型。首先通过辅助的预检测模型估计观测平面上小目标可能出现的区域,然后根据该区域初始化粒子分布,替代常用均匀分布于观测平面的粒子初始化分布,用少量的粒子对低信噪比光电图像序列中的小目标进行跟踪。实验表明,与Salmond提出的粒子滤波目标跟踪算法相比,该方法跟踪精度指标均方根误差从5.29下降到0.38,算法CPU执行时间从0.83 s下降到0.25 s,算法的复杂度下降了69.8%,满足了对算法实时性的需求。

The problem of tracking small target＇s high computation complexity in optical image sequence using a single sensor was considered in this paper. To cut down the computational complexity of particle filter for target tracking, an improved particle filter for small target tracking based on target region pre-detect model was proposed. The main improvement was the pre-detect model which is the basement of shrinking particles＇ initial prior distribution. First, the pre-detect model was adopted to identify the approximate target region. Then the conventional uniform distribution of particles over the whole sensor field-of-view is replaced by the pre-detection target area. Therefore, the target state was represented by fewer particles effectively. To demonstrate the efficiency of the proposed method, it was compared with the particle filter proposed by Salmond. Simulation results indicate that the Root Mean Squared Error （RMSE）, the criterion of tracking accuracy, reduces from 5.29 to 0.38, the CPU time reduces from 0.83 seconds to 0.25 seconds which means the 80% computational complexity decreasing. The reduction of computational complexity satisfies the real-time demand of the proposed algorithm.