摘要
由于跟踪器采用的是基于金字塔Lucaks-Kanade光流法,需要在相邻帧之间对目标的运动进行跟踪,运算量较大,因此提出了采用轻量级跟踪算法Camshift作为TLD算法框架中的跟踪器模块,来提高跟踪模块运行效率。而TLD框架的检测器在跟踪器追踪失败时需要检测大量数目的子窗口,因此利用背景差分方法进行前景检测,可以减小检测范围和数目。TLD算法本身是对单目标的长时间跟踪,提出基于多线程机制TLD算法,针对每一个跟踪目标建立相应的线程对其跟踪。经过实验验证,与原算法相比,优化算法使得对多目标实时跟踪性能得到一定提升。
The tracker in tracking learning detection（ TLD） framework is based on the pyramid Lucaks-Kanade optical flow method,which needs a large amount of computation between adjacent frames for target tracking,and therefore,this paper proposed an optimized tracking algorithm based on lightweight Camshift as the TLD tracker module to improve the tracker efficiency. The TLD frame detector needs to check a large amount of sub-windows when the tracker fails to track object,and thus foreground detection was proposed to decrease the detection range and number by making use of background differentiation method. TLD itself is a long term tracking framework for a single target. This paper put forward TLD based on multithreading programming to establish the corresponding thread that is responsible for tracking its target. Related experimental results show that the performance of optimized algorithm for tracking multiple targets gets better improvement.
出处
《应用科技》
CAS
2016年第5期59-64,共6页
Applied Science and Technology
