基于椭圆RHM的扩展目标伯努利滤波算法

与传统点目标跟踪不同,扩展目标跟踪既要估计目标的运动状态,还需估计目标的扩展状态,包括目标的形状、大小、方向等信息。针对扩展目标跟踪中存在的扩展状态估计不准确和非线性问题,提出一种基于随机超曲面模型(random hypersurface model, RHM)的扩展目标伯努利滤波算法。该算法首先采用RHM对目标量测源建模;然后,在扩展目标伯努利滤波框架下,实现对单扩展目标运动状态和扩展状态的实时估计;最后,引入Gamma分布以提高量测率估计的准确性。此外,为了降低计算复杂度,在量测更新中采用距离划分来减少所有可能的划分总数。实验结果表明,所提滤波算法在估计目标运动状态、扩展状态和量测率等方面优于现有的滤波算法,并且可用于实际视频跟踪场景。

Extended-target Bernoulli filter algorithm based on ellipse RHM

Different from the traditional point target tracking, the extended-target tracking not only estimates the target’s kinematical state, but also estimates the target’s extension state, including target’s shape, size and orientation. Aimed at the problem of the inaccurate and nonlinear estimation of the target shape, an algorithm based on random hypersurface model (RHM) and Bernoulli filter is proposed. Firstly, the measurement source of target is modeled as an ellipse random hypersurface model, and then it is embedded into the Bernoulli filter to track an extend target in real time. Finally, Gamma distribution is applied into the filter to estimate the measurement rate for improving the accuracy of estimation. A cluster step is inserted into the measurement update using the distance partitioning method in order to reduce the computational complexity. Experiment result shows that the proposed algorithm outperforms the traditional Bernoulli filter in estimating the trajectory and shape of the target, and can be used for practical video tracking scenarios.