基于粒子滤波的红外多目标跟踪算法研究

发布时间：2018-09-14 20:17

【摘要】：近几年,随着信息技术的快速发展,尤其是计算技术和红外成像技术的飞速发展,红外多目标检测与跟踪技术在军事和民用中都起着更加重要的作用。尤其是在跟踪多目标时,在多目标被遮挡以及跟踪的实时性等问题成为研究的热点也是难点,存在着许多问题需要解决或改进。这些年以来,粒子滤波算法作为当代非线性滤波最为重要的一类滤波方法,已经取得了非常大的进步。首先,本文分析了经典粒子滤波算法的原理以及经典粒子滤波算法在红外多目标跟踪中的应用,并且对多目标跟踪的相关问题进行了数学建模。接下来在贝叶斯跟踪理论框架下分析讨论了该结构的优缺点。其次,描述了均值漂移算法的一些简单理论,然后提出了一种将均值漂移算法与粒子滤波跟踪算法相结合使用的红外目标跟踪方法。这种方法保持了均值漂移算法计算量小,并且具有很好的实时性的特点。通过均值漂移算法对粒子滤波中的粒子进行收敛,使每个粒子都具有更真实的目标特性,大大减少了描述目标状态所需要的粒子数目,并且提升了粒子的采样效率,提高了算法的实时性。通过几组实验证明,融合后的算法具有稳健的跟踪性能,并且节省了计算时间兼顾到了目标跟踪中的实时性要求。最后,本文研究了改进的粒子滤波多目标检测算法。介绍了一种常见的红外多目标检测算法;在分析其本身的优缺点基础上,提出一种基于权值优选的红外多目标检测算法,通过改进重采样方式,使得权值较大的粒子进入下一步跟踪而舍弃权值较小的粒子,提升了跟踪的准确性以及持久性。随后,在基于权值优选粒子滤波算法中引入了马尔可夫(Markov)随机场,用无向图模型来处理多目标跟踪时的数据关联问题,使得该算法可以对遮挡多目标进行跟踪,提高了目标被部分遮挡时的跟踪的有效性。本论文研究的改进粒子滤波算法在跟踪的实时性和抗遮挡方面较常见的粒子滤波算法有很大的改进,对多目标跟踪技术方面有一定的理论意义和应用价值。
[Abstract]:In recent years, with the rapid development of information technology, especially the rapid development of computing technology and infrared imaging technology, infrared multi-target detection and tracking technology plays a more important role in both military and civilian applications. Especially when tracking multiple targets, the problems of occlusion and real-time tracking become hot and difficult, and many problems need to be solved or improved. In recent years, particle filter algorithm, as one of the most important filtering methods for nonlinear filtering, has made great progress. Firstly, the principle of classical particle filter algorithm and the application of classical particle filter algorithm in infrared multi-target tracking are analyzed, and the related problems of multi-target tracking are modeled. Then, the advantages and disadvantages of the structure are analyzed and discussed in the framework of Bayesian tracking theory. Secondly, some simple theories of mean shift algorithm are described, and then an infrared target tracking method which combines mean shift algorithm with particle filter tracking algorithm is proposed. This method keeps the calculation complexity of mean shift algorithm small, and has the characteristic of good real-time performance. The mean shift algorithm is used to converge the particles in the particle filter, which makes each particle have more real target characteristics, greatly reduces the number of particles needed to describe the target state, and improves the sampling efficiency of the particle. The real-time performance of the algorithm is improved. Several experiments show that the fusion algorithm has robust tracking performance and saves computational time to meet the real-time requirements of target tracking. Finally, the improved particle filter multi-target detection algorithm is studied. This paper introduces a common infrared multi-target detection algorithm, based on the analysis of its own merits and demerits, proposes an infrared multi-target detection algorithm based on weight optimal selection, and improves the method of resampling. The higher weight particle goes into the next step of tracking and the smaller weight particle is abandoned, which improves the accuracy and persistence of the tracking. Then, the Markov (Markov) random field is introduced into the weighted optimal particle filter algorithm, and the undirected graph model is used to deal with the data association problem in the multi-target tracking, so that the algorithm can track the occluded multi-target. The effectiveness of tracking is improved when the target is partially occluded. The improved particle filter algorithm studied in this paper has a great improvement in the real-time and anti-occlusion aspects of tracking, and has a certain theoretical significance and application value to the multi-target tracking technology.
【学位授予单位】：沈阳理工大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.41

【参考文献】