摘要
TLD(Tracking—Learning—Detection)是近期受到广泛关注的一种有效的视频目标跟踪算法。基于经典的TLD算法,通过构建协同跟踪框架,将待跟踪目标的目标位置信息和特征信息加入到TLD算法的学习模型中,实现了利用多无人机协同跟踪同一或多个运动目标。利用TLD算法进行多无人机协同目标跟踪,算法的跟踪准确率最低为93.9%,最高可达到99.1%,且每帧的平均跟踪时间在41~47ms之间,跟踪性能较好,基本可以达到实时的性能;另外,跟踪算法的平均像素偏移数最大仅为5.013个像素,跟踪精度高。结果表明,利用TLD算法进行多无人机协同目标跟踪是可行的,且具有一定的应用和研究价值。
TLD(Tracking-Leaming-Detection) is recently an effective video object tracking algorithm with widespread attention. The single or multiple moving targets collaborative tracking is achieved by using muhi-UAV based on the classic TLD algorithm, through building collaborative tracking framework and adding the target position information and feature information to the learning model of TLD algorithm. Experimental results show that the algorithm accuracy is from 93. 9% to 99. 1 %, and the average tracking time per frame is between 41-47 ms, the tracking performance is better. Further, the largest average pixel shift of the algorithm is only 5. 013 pixels to prove that it has high accuracy. Finally, multi-UAV collaborative target tracking technology based on TLD algorithm is viable.
出处
《光学与光电技术》
2015年第5期82-86,共5页
Optics & Optoelectronic Technology
关键词
目标跟踪
TLD算法
协同跟踪框架
学习模型
平均跟踪时间
平均像素偏移数
target tracking
TLD algorithm
collaborative tracking framework
learning model
average tracking time
average shift number of pixels