基于序贯蒙特卡罗概率假设密度滤波的多目标检测前跟踪改进算法

Improved Multitarget Track Before Detect Algorithm Using the Sequential Monte Carlo Probability Hypothesis Density Filter

实现目标数目未知且可变条件下的多目标检测与跟踪是个极具挑战性的问题,在信噪比较低的情况下更是如此。针对这一问题,该文提出一种基于点扩散模型的多目标检测前跟踪改进算法。该算法在序贯蒙特卡罗概率假设密度(SMC-PHD)滤波框架下实现,通过自适应粒子产生机制完成新生目标在像平面中的初始定位,并根据目标在图像中可能出现的位置对全体粒子集进行有效子集分割和快速权值估算,最后利用动态聚类方法完成多目标状态的准确提取。仿真结果表明,该方法有效改善了多目标检测前跟踪的估计性能,并大大提高了算法执行效率。

The Detection and tracking of multi-target is a challenging issue under the condition with unknown and varied target number, especially when the Signal-to-Noise Ratio （SNR） is low. An improved Track-Before-Detect （TBD） method for multiple spread targets is proposed by using point spread observation model. The method is prepared from the framework of the Sequential Monte Carlo Probability Hypothesis Density （SMC-PHD） filter, and it is implemented by firstly adopting an adaptive particle generation strategy, which can obtain the rough position estimates of the potential targets. The particle set is then partitioned into multiple subsets according to their position coordinates in 2D image plane and an efficient evaluation of the updated particle weights is accomplished by utilizing the convergence property of the particles. Target tracks are finally constructed from the extracted multitarget states via dynamic clustering technique. Simulation results show that the presented method can not only greatly improve the performance of multitarget TBD, but also significantly reduce the executing time of SMC-PHD based implementation.