贝叶斯框架下多传感器目标跟踪算法研究

发布时间：2018-05-29 02:05

本文选题：目标跟踪 + 定位　；参考：《浙江大学》2017年硕士论文

【摘要】：随着网络通信技术的不断发展,由于多传感器观测通常会削减估计的不确定性,现代防御观测系统逐渐向多基地化和网络化发展。在多传感器观测系统中,如何进行数据融合是一个十分重要的问题,尤其是伴随着传感器工艺的不断提升,越来越多的系统使用低成本的传感器进行较大规模的组网观测,经典的方法在没有融合中心的情形下具有一定的局限性;此外,在实际的观测过程中,时常会伴随着检测的不确定性,因此导致虚警,这使得经典的贝叶斯滤波算法在上述情形中具有局限性;为了解决上述问题,本文在这两方面展开了较为深入的研究。特别的,本文将目标跟踪问题归类为良好检测条件下和非良好检测条件下的状态估计问题。在良好检测条件下,本文研究了基于后验概率密度一致性的分布式估计算法,并分析了协方差嵌入数据融合和基于信息散度的一致性估计之间的联系,在纯方位跟踪情形下的数值分析证明了该算法的有效性。在非良好检测条件下,本文在随机有限集框架下研究了多传感器目标跟踪问题,利用序贯似然函数更新,实现了多传感器伯努利粒子滤波,并以方位观测为模型,进行了理论分析。数值仿真结果证明了该算法的有效性。最后,本文对于基于标记随机有限集的多目标跟踪算法进行了研究。数值仿真结果表明该算法能够同步标记以及估计多个目标的状态。
[Abstract]:With the continuous development of network communication technology, because the uncertainty of estimation is usually reduced by multi-sensor observation, the modern defense observation system is gradually becoming multi-static and networked. In the multi-sensor observation system, how to fuse data is a very important problem, especially with the continuous improvement of sensor technology, more and more systems use low-cost sensors to conduct large-scale network observation. The classical method has some limitations in the absence of fusion center. In addition, in the actual observation process, the uncertainty of detection is often accompanied, which leads to false alarm. This makes the classical Bayesian filtering algorithm have limitations in the above cases. In order to solve the above problems, this paper has carried out a more in-depth study in these two aspects. In particular, the target tracking problem is classified as the state estimation problem under the condition of good detection and non-good detection. In this paper, a distributed estimation algorithm based on posteriori probability density consistency is studied under good detection conditions, and the relationship between covariance embedded data fusion and consistency estimation based on information divergence is analyzed. The effectiveness of the algorithm is proved by numerical analysis in the case of azimuth-only tracking. In this paper, the problem of multi-sensor target tracking is studied under the frame of random finite set under the condition of unfavorable detection. The Bernoulli particle filter of multi-sensor is realized by updating the sequential likelihood function, and the azimuth observation model is used as the model. Theoretical analysis is carried out. Numerical simulation results show the effectiveness of the algorithm. Finally, this paper studies the multi-target tracking algorithm based on labeled random finite set. Numerical simulation results show that the algorithm can synchronously mark and estimate the state of multiple targets.
【学位授予单位】：浙江大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TN713;TP212

【参考文献】