摘要
TLD目标跟踪算法将检测和跟踪同时加入跟踪框架,并引入半监督机器学习算法对锁定目标不断学习以捕获其最新外观,使目标无处可逃,从而实现对未知目标的长时跟踪。简要介绍了TLD算法的理论背景和系统框架,并指出其存在的主要缺陷。然后对各个缺陷所提出的改进方法进行详细地综述,并给出自己的评述。最后总结全文并展望TLD目标跟踪算法在未来几年的发展趋势。
In order to enable long-term tracking for unknown objects, a semi-supervized machine learning algorithm is added in ob- ject tracking based on TLD algorithm which matches traditional tracking algorithm and detection algorithm for keeping capturing the newest appearances by continuous learning. Firstly, TLD＇ s theoretical background, tracking framework and major defects are in- troduced. Then, the improvements of the original algorithm are elaborately disscussed. Finally, the propspects of object tracking based on TLD algorithm are presented.
出处
《电视技术》
北大核心
2016年第10期101-106,共6页
Video Engineering
基金
2015年重庆市研究生科研创新项目(CYS15166)
