相关滤波目标跟踪进展综述

发布时间：2018-04-11 02:19

本文选题：相关滤波跟踪 + 特征表示　；参考：《中国图象图形学报》2017年08期

【摘要】：目的目标跟踪是计算机视觉中的关键问题,在人机交互、行为识别等领域有着非常广泛的应用。最近,相关滤波理论由于其高效性和鲁棒性,被用于目标跟踪领域,取得了一系列新的进展,成为目标跟踪领域的研究热点,得到越来越多的关注。为使更多国内外研究者对相关滤波目标跟踪理论及其发展进行探索,本文对该领域研究现状进行综述。方法首先介绍相关滤波跟踪的一般框架,在此基础上,给出相关滤波理论,并重点对经典相关滤波跟踪算法——核相关滤波(KCF)跟踪算法进行详细描述,然后讨论目标跟踪中常见问题的处理——特征的有效表示和尺度变化的适应,并进一步从3个方面分析当前研究现状,指出未来可能的发展趋势。结果采用目标跟踪基准数据库(OTB-2013)中的50组视频序列进行实验,分析比较45种不同跟踪算法的性能,包括14种代表性相关滤波跟踪算法。在总体性能评估中给出排名前15位的跟踪算法,其中相关滤波跟踪算法占11个,充分体现此类算法的优越性。结论相关滤波理论在目标跟踪领域的研究取得了显著的进展,具有广阔的应用前景,但受复杂场景及目标外观剧烈变化等因素影响,目标跟踪仍是一个极具挑战的问题。研究高效、鲁棒的相关滤波目标跟踪算法具有非常重要的意义。
[Abstract]:Objective Target tracking is a key problem in computer vision. It is widely used in human-computer interaction and behavior recognition.Recently, because of its high efficiency and robustness, the related filtering theory has been used in the field of target tracking. It has made a series of new progress, and has become a research hotspot in the field of target tracking, which has attracted more and more attention.In order to make more researchers at home and abroad to explore the relevant filtering target tracking theory and its development, this paper summarizes the current research situation in this field.Methods first of all, the general framework of correlation filter tracking is introduced, and then the theory of correlation filter is given, and the classical correlation filter tracking algorithm, kernel correlation filter tracking algorithm, is described in detail.Then, the effective representation of common problems in target tracking and the adaptation of scale variation are discussed, and the current research situation is analyzed from three aspects, and the possible development trend in the future is pointed out.Results A total of 50 groups of video sequences from OTB-2013) were used to analyze and compare the performance of 45 different tracking algorithms, including 14 representative correlation filtering and tracking algorithms.In the overall performance evaluation, the top 15 tracking algorithms are given, among which 11 are related filtering tracking algorithms, which fully reflect the advantages of this algorithm.Conclusion the related filtering theory has made remarkable progress in the field of target tracking and has a broad application prospect. However, the target tracking is still a very challenging problem due to the influence of complex scenes and dramatic changes in the appearance of the target.It is very important to study the efficient and robust correlation filtering target tracking algorithm.
【作者单位】：西北大学信息科学与技术学院;
【基金】：西北大学研究生自主创新基金项目(YZZ15094)
【分类号】：TP391.41

【相似文献】