基于形状估计的随机集多扩展目标跟踪方法研究

发布时间：2018-03-20 07:37

本文选题：随机有限集　切入点：扩展目标　出处：《西安电子科技大学》2015年硕士论文　论文类型：学位论文

【摘要】：由于雷达和传感器的分辨率随科技进步而不断提高,得到的同一个目标的量测不止一个,此时,目标需看作是扩展目标,如果仍然使用传统的方法将量测和目标相关联进行跟踪已无法满足现状。近些年,基于随机有限集的多目标跟踪方法由于能够有效地解决传统跟踪方法中出现的一些难题,并且在计算复杂度方面明显优于传统算法,所以受到了广泛学者的认同。本文主要针对扩展目标中基于形状估计的随机集多目标跟踪方法展开研究。主要研究内容如下:1.针对K-means聚类划分方法过分依赖初始中心点以及效率低的问题,提出了一种基于均值漂移聚类的改进划分方法。该方法将高斯函数作为内核函数,采用均值漂移方法求出吸引盆,并将吸引盆进行合并,最后将噪声孤点去除,得到最终的量测划分集合。该方法不需要给出具体的聚类数目和中心点,可得到良好的划分效果。2.针对距离划分方法在目标相邻时无法进行准确划分的问题,提出了一种基于距离划分的改进方法。该方法在距离划分的基础上进行二次划分,使用极大似然估计来估计每个划分集合中的目标数,对目标数大于1的划分集合进行再划分,从而将每个扩展目标的量测划分开。该方法可以有效地处理目标相邻或交叉时距离划分无法区分的情况。3.针对现有扩展目标的建模使用椭圆建模而无法分辨星形形状的问题,提出了一种基于星凸型随机超曲面的伽玛高斯混合势概率假设密度扩展目标跟踪算法(SRHM-GGM-CPHD)。该算法将扩展目标的形状建模为星凸形,并将其嵌入到伽玛高斯混合CPHD滤波器框架中,完成对多个扩展目标的跟踪。该算法在质心位置和扩展形状的估计精度方面要优于传统的基于随机矩阵的伽玛高斯逆威舍特CPHD滤波器。
[Abstract]:Since the resolution of radar and sensors has improved with technological advances, more than one measurement of the same target has been obtained, where the target needs to be viewed as an extended target. In recent years, the multi-target tracking method based on stochastic finite set can effectively solve some difficult problems in traditional tracking methods. And the computational complexity is obviously better than the traditional algorithm, This paper mainly focuses on the shape estimation based random set multi-target tracking method in extended targets. The main contents of the research are as follows: 1. K-means clustering method is too dependent on the initial. The center point and the problem of inefficiency, An improved partition method based on mean shift clustering is proposed, in which Gao Si function is taken as kernel function, mean shift method is used to calculate the suction basin, and the suction basin is merged, finally the noise is removed from the isolated point. The final measurement partition set is obtained. This method does not need to give specific clustering number and center point, and can get good partition effect. 2. Aiming at the problem that the distance partition method can not be accurately partitioned when the target is adjacent, An improved method based on distance partition is proposed, in which the quadratic partition is based on the distance partition. The maximum likelihood estimation is used to estimate the number of objects in each partition set, and the partition set with more than one target number is repartitioned. Thus, the measurement of each extended target can be divided. The method can effectively deal with the problem that the distance partition between adjacent or crossing targets can not be distinguished. 3. For the existing extended target modeling, the elliptical model is used but the star shape can not be distinguished. In this paper, a hybrid probability assumption density extended target tracking algorithm for gamma Gao Si based on star convex random hypersurface is proposed. The extended target shape is modeled as a star convex shape by this algorithm and embedded in the framework of Gamma Gao Si hybrid CPHD filter. The algorithm is superior to the conventional Gamma Gao Si inverse Weschet CPHD filter based on random matrix in the estimation accuracy of centroid position and extended shape.
【学位授予单位】：西安电子科技大学
【学位级别】：硕士
【学位授予年份】：2015
【分类号】：TN713

【参考文献】