摘要
为提升TLD目标跟踪算法的每帧处理速度,以达到在更高分辨率视频中跟踪目标的实时性要求,在TLD算法框架的基础上,提出了一种基于自适应尺度检测学习的目标跟踪算法(AS-TLD)。当跟踪目标成功时,选取当前帧跟踪到的目标尺度及几个相邻的尺度作为下帧检测目标时滑动窗口尺度的选取范围;而当跟踪失败时,则选取在TLD算法初始化阶段,根据跟踪目标及视频图像大小选定的尺度来保障长时间跟踪目标,从而有效减少了平均每帧扫描的窗口数量。实验结果表明,该方法不仅有效地降低了检测模块的检测时间,显著提高了整体算法速度,而且通过动态选取尺度,在一定程度使得TLD各个模块更加协调,跟踪精确度得到提升。
In order to process video image with high frame rate and meet the real-time requirement in higher resolution,adaptive choosing scales target tracking method(AS-TLD)is proposed,based on TLD algorithm frame.When tracking target successfully,the target scale in current frame and several adjacent scales are selected as the sliding window scales range in the next frame.While tracking algorithm is disabled,those scales defined at algorithm initialization phase are chosen by the frame image size and selected box size.Therefore,the algorithm can effectively reduce average number of scanning window.The experiment results show that the proposed approach reduces detection module running time effectively,as well increase overall algorithm speed.And by using the adaptive scale method,TLD modules are more harmonious,tracking accuracy is improved too in some degree.
出处
《光学技术》
CAS
CSCD
北大核心
2017年第6期542-546,共5页
Optical Technique
基金
广东省科技计划项目(2013B010204055)资助