基于加速鲁棒特征和多示例学习的目标跟踪算法

发布时间：2018-04-10 11:41

  本文选题：加速鲁棒特征 + 多示例学习　； 参考：《计算机应用》2016年11期

【摘要】：针对照明变化、形状变化、外观变化和遮挡对目标跟踪的影响,提出一种基于加速鲁棒特征(SURF)和多示例学习(MIL)的目标跟踪算法。首先,提取目标及其周围图像的SURF特征;然后,将SURF描述子引入到MIL中作为正负包中的示例;其次,将提取到的所有SURF特征采用聚类算法实现聚类,建立视觉词汇表;再次,通过计算视觉字在多示例包的重要程度,建立"词-文档"矩阵,并且求出包的潜在语义特征通过潜在语义分析(LSA);最后,通过包的潜在语义特征训练支持向量机(SVM),使得MIL问题可以依照有监督学习问题进行解决,进而判断是否为感兴趣目标,最终实现视觉跟踪的目的。通过实验,明确了所提算法对于目标的尺度缩放以及短时局部遮挡的情况都有一定的鲁棒性。
[Abstract]:Aiming at the influence of illumination change, shape change, appearance change and occlusion on target tracking, a target tracking algorithm based on accelerated robust feature tracking (surf) and multi-example learning algorithm (MIL) is proposed.Firstly, the SURF features of the target and its surrounding images are extracted; then, the SURF descriptor is introduced into the MIL as an example of positive and negative packets. Secondly, all the extracted SURF features are clustered by clustering algorithm to establish the visual vocabulary.By calculating the importance of visual words in multi-sample packets, the "word-document" matrix is established, and the potential semantic features of the packets are obtained through potential semantic analysis.By training support vector machines with the potential semantic features of packets, the MIL problem can be solved according to supervised learning problems, and then determine whether it is the object of interest, and finally achieve the purpose of visual tracking.Through experiments, it is clear that the proposed algorithm is robust to the scale scaling of the target and the local occlusion in a short time.
【作者单位】： 山西大学计算机与信息技术学院;西安工程大学计算机科学学院;
【基金】：国家自然科学基金资助项目(61201453,61201118) 山西省基础研究计划项目(2014021022-2) 山西省高等学校科技创新项目(2015108)~~
【分类号】：TP391.41
，

本文编号：1731053