基于子空间半监督学习线性判别方法的目标跟踪技术研究
|
摘要
文中提出一种基于半监督学习的线性判别方法用于目标跟踪。首先,根据少量的目标图像和背景图像样本,利用增量线性判别分析在子空间中找到最大化标记样本分类间隔的分类面;然后在当前帧采样,获得大量未标记的图像样本并投影到子空间中,通过半监督学习修正分类面,在这些候选目标中找到离目标最近、离背景最远的作为目标在当前帧的状态估计;最后,在分类结果中挑选置信度高的目标图像和背景图像样本加入到训练集中,删除训练集中置信度低的目标图像和背景图像样本,并更新投影子空间的基。实验结果表明,所提方法可以很好地适应目标的各种变化,并获得比基于监督学习方法更好的效果。
A linear discriminate analysis method based on semi-supervised learning is proposed for object tracking.According to the few object image and background image samples,the incremental linear discriminate analysis is used to find the classification plane with maximum labeled sample classification interval. The current frame is sampled to acquire a large number of unlabeled image samples,and the samples are projected into subspace. The semi-supervised learning is used to modify the classification plane. The nearest object which is farthest away the background in candidate objects is selected as the target to estimate the state of current frame. The object image and background image samples with high confidence are selected from classification results to add them into the training set. The object image and background image samples with low confidence are deleted in training set,and the bases of projected subspace are updated. The experimental results show that the proposed method can adapt to various changes of object perfectly,and acquire better results than the method based on supervised learning.
A linear discriminate analysis method based on semi-supervised learning is proposed for object tracking.According to the few object image and background image samples,the incremental linear discriminate analysis is used to find the classification plane with maximum labeled sample classification interval. The current frame is sampled to acquire a large number of unlabeled image samples,and the samples are projected into subspace. The semi-supervised learning is used to modify the classification plane. The nearest object which is farthest away the background in candidate objects is selected as the target to estimate the state of current frame. The object image and background image samples with high confidence are selected from classification results to add them into the training set. The object image and background image samples with low confidence are deleted in training set,and the bases of projected subspace are updated. The experimental results show that the proposed method can adapt to various changes of object perfectly,and acquire better results than the method based on supervised learning.
引文
[1]杨晓伟,赵海军.自动目标识别与机动目标建模及跟踪方法研究[J].电子科技,2017,30(10):127-130.YANG Xiaowei,ZHAO Haijun. Research on automatic target recognition and maneuvering target modeling and tracking[J].Journal of electronic science and technology,2017,30(10):127-130.
[2]朱文青,刘艳,卞乐,等.基于生成式模型的目标跟踪方式综述[J].微处理机,2017(1):41-47.ZHU Wenqing,LIY Yan,BIAN Le,et al. Survey on object tracking method base on generative model[J]. Microprocessors,2017(1):41-47.
[3]王苹,张磊,吴泽民,等.视频目标跟踪典型算法比较与分析[J].电视技术,2015,39(23):79-84.WANG Ping,ZHANG Lei,WU Zemin,et al. Comparison and analysis of state-of-the-art visual object tracking algorithms[J].Video engineering,2015,39(23):79-84.
[4]吴建宁,徐海东,凌云,等.基于块稀疏贝叶斯学习的人体运动模式识别[J].计算机应用,2016,36(4):1039-1044.WU Jianning,XU Haidong,LING Yun,et al. Human activity pattern recognition based on block sparse Bayesian learning[J].Journal of computer applications,2016,36(4):1039-1044.
[5]张汝云,许明艳,江涛.一种基于隐马尔科夫模型的目标轨迹跟踪算法[J].计算机应用研究,2013,30(5):1455-1470.ZHANG Ruyun,XU Mingyan,JIANG Tao. Partial CSI based affinity propagation dynamic clustering algorithm[J]. Application research of computers,2013,30(5):1455-1470.
[6] AKARM B, MARC P, DUC A T. A study on human activity recognition using accelerometer data from smart phones[J].Procedia computer science,2014,34:450-457.
[7] ZHANG X,HU W,MAYBANK S,et al. Graph based discriminative learning for robust and efficient object tracking[C]//2007 IEEE the 11th International Conference on Computer Vision. Rio de Janeiro,Brazil:IEEE,2007:1-8.
[2]朱文青,刘艳,卞乐,等.基于生成式模型的目标跟踪方式综述[J].微处理机,2017(1):41-47.ZHU Wenqing,LIY Yan,BIAN Le,et al. Survey on object tracking method base on generative model[J]. Microprocessors,2017(1):41-47.
[3]王苹,张磊,吴泽民,等.视频目标跟踪典型算法比较与分析[J].电视技术,2015,39(23):79-84.WANG Ping,ZHANG Lei,WU Zemin,et al. Comparison and analysis of state-of-the-art visual object tracking algorithms[J].Video engineering,2015,39(23):79-84.
[4]吴建宁,徐海东,凌云,等.基于块稀疏贝叶斯学习的人体运动模式识别[J].计算机应用,2016,36(4):1039-1044.WU Jianning,XU Haidong,LING Yun,et al. Human activity pattern recognition based on block sparse Bayesian learning[J].Journal of computer applications,2016,36(4):1039-1044.
[5]张汝云,许明艳,江涛.一种基于隐马尔科夫模型的目标轨迹跟踪算法[J].计算机应用研究,2013,30(5):1455-1470.ZHANG Ruyun,XU Mingyan,JIANG Tao. Partial CSI based affinity propagation dynamic clustering algorithm[J]. Application research of computers,2013,30(5):1455-1470.
[6] AKARM B, MARC P, DUC A T. A study on human activity recognition using accelerometer data from smart phones[J].Procedia computer science,2014,34:450-457.
[7] ZHANG X,HU W,MAYBANK S,et al. Graph based discriminative learning for robust and efficient object tracking[C]//2007 IEEE the 11th International Conference on Computer Vision. Rio de Janeiro,Brazil:IEEE,2007:1-8.