计算机工程与应用 ›› 2019, Vol. 55 ›› Issue (12): 124-131.DOI: 10.3778/j.issn.1002-8331.1811-0391

• 模式识别与人工智能 • 上一篇    下一篇

高斯混合模型结合加权似然的目标跟踪算法

陈  超   

  1. 内江师范学院 四川省数据恢复重点实验室,四川 内江 641110
  • 出版日期:2019-06-15 发布日期:2019-06-13

Target Tracking Algorithm Involving Gaussian Mixture Model and Weighted Likelihood

CHEN Chao   

  1. Data Recovery Key Laboratory of Sichuan Province, Neijiang Normal University, Neijiang, Sichuan 641110, China
  • Online:2019-06-15 Published:2019-06-13

摘要: 鉴于高斯混合模型对背景变化快时无法精确检测出目标和目标跟踪的适应性差等瑕疵,提出了基于加权似然跟踪器来改进高斯混合模型实现运动目标跟踪算法。主要引入了自适应高斯混合模型来实时检测运动目标,然后空间加权似然来进行视频中的目标定位,引入加权似然期望值来改进高斯混合模型处理视频中的多尺度、多角度变化的目标跟踪不精准问题。通过VOT 2014 dataset对比实验结果表明提出的基于加权似然跟踪(Weighted Likelihood Tracking,WLT)和改进高斯混合模型(Improved Gaussian Mixture Model,IGMM)的目标跟踪算法较传统高斯混合模型跟踪算法在跟踪的精度有较大提高。在应对多尺度、多角度变化的目标跟踪表现出了较大的优势。

关键词: 改进高斯混合模型, 分数阶导数学习率, 目标跟踪算法, 加权似然跟踪, 期望值最大化

Abstract: The target tracking algorithm based on WLT and IGMM is proposed in view of the GMM algorithm’s poor adaptability while the background changes fastly and target are multiple. An adaptive Gauss mixture model involving fractional derivative learning rate is introduced to detect moving targets in real time. The target localization is achieved by maximizing its weighted likelihood in the video. Moreover, the algorithm handles scale and rotation changes of the mulitarget. Experimental results in VOT2014 dataset suggested proposed algorithm involving WLT and IGMM comparing current popular tracking algorithm in tracking accuracy is improved greatly. In respond to the changes in multi-scale, multi-angle multi-target tracking shows greater advantage.

Key words: Improved Gaussian Mixture Model(IGMM), fractional derivative learning rate, target tracking algorithm, Weighted Likelihood Tracking(WLT), expectation-maximization