未知测量噪声分布下的多目标跟踪算法

 粒子概率假设密度滤波(SMC-PHDF)在进行粒子更新时需要知道测量噪声的概率分布以计算似然函数,这使得SMC-PHDF依赖于测量噪声的概率模型。针对这一点不足,提出一种未知测量噪声分布下的多目标跟踪算法——基于风险评估的概率假设密度滤波(RE-PHDF)。该算法在SMC-PHDF进行概率假设密度(PHD)粒子更新时采用风险函数计算每个PHD粒子的风险值,并通过一个风险评估函数评估每个PHD粒子,然后用评估后的结果更新粒子的权值。由于粒子更新时避免了在多维测量空间中计算似然函数,算法不仅不依赖于测量噪声的概率分布,还可以节省大量计算时间。仿真结果表明：和SMC-PHDF相比,RE-PHDF在未知的复杂测量噪声环境下具有更高的鲁棒性和稳定性;同时,在两种算法跟踪精度接近的情况下,所提算法节省了50%的运行时间。

A Multi-target Tracking Algorithm Under Unknown Measurement Noise Distribution

When updating particles, a particle probability hypothesis density filter (SMC-PHDF) requires the probabilistic distribution of measurement noise to calculate the likelihood function, which makes it rely excessively on the probabilistic model of measurement noise. To overcome this drawback, a new multiple target tracking algorithm under unknown probabilistic distribution of measurement noise is proposed, namely, a risk evaluation-based probability hypothesis density filter (RE-PHDF). When SMC-PHDF updates probability hypothesis density(PHD) particles, the algorithm computes the risk of each particle using a risk function, and evaluates each particle by a risk evaluation function, then updates the particle weights by means of the evaluated results. Avoiding thus the likelihood function calculation in multi-dimensional measurement space, the algorithm does not depend on the probabilistic distribution of measurement noise and can save much computing time. The simulation results show that RE-PHDF possesses higher robustness and stability under unknown and complicated measurement noise environment in comparison with SMC-PHDF. In addition, the new algorithm can save up to 50% execution time while possessing similar accuracy as SMC-PHDF.