一种改进的TLD跟踪算法研究  

Research on An Improved TLD Tracking Algorithm

作　　者：徐攀[1] 齐文宗[2] XU Pan;QI Wen-zong(Jinjiang College,Sichuan University,Meishan Sichuan 620800,China;School of Electronic Information Sichuan University,Chengdu Sichuan,610000,China)

机构地区：[1]四川大学锦江学院,四川眉山620800 [2]四川大学电子信息学院,四川成都610000

出　　处：《计算机仿真》2022年第4期312-315,共4页Computer Simulation

摘　　要：在TLD(Tracking-Learning-Detection)跟踪算法的研究上,提出了一种基于SVM(Support Vector Machine)的TLD目标跟踪算法。改进的TLD跟踪算法采用支持向量机(SVM)分类器进行图像目标正负样本的分类学习,有效提高了算法的鲁棒性以及实时性。另外对算法的Haar-Like特征进行了改进,利用多种Haar-Like特征,能提取目标丰富的特征信息。研究结果表明,改进算法能有效的长时间的跟踪目标,减少了目标在跟踪过程中的漂移现象,提高了跟踪的鲁棒性的准确性。On the basis of TLD tracking algorithm, a TLD target tracking algorithm based on SVM algorithm is proposed in this paper. The improved TLD tracking algorithm uses support vector machine(SVM) classifier to image target classification learning of positive and negative samples, and effectively improved the robust of the algorithm and the real-time performance. Moreover, the algorithm of Haar-Like characteristics was improved in this paper. By using a variety of Haar-Like features, it can extract rich feature information of target. Research results show that the improved algorithm can effectively track target for a long period of time, reduce the target drift phenomenon in the process of tracking, and improve the accuracy of the tracking robustness in this paper.

关 键 词：目标跟踪 检测器 支持向量机 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]