摘要
为实现在复杂背景和多干扰条件下红外目标的稳定跟踪,提出一种基于多特征相关滤波的红外目标跟踪算法。首先综合考虑生物视觉关注特性及目标运动特性,提取目标区域的空间特征和运动特征,进而融合一种改进的卷积特征,生成多特征权值函数;然后在传统相关滤波的基础上,引入多特征权值函数用以表征不同候选区域的重要程度,形成权值相关滤波的红外目标跟踪框架;最终得到能够表征目标位置的置信图,从而完成红外目标的鲁棒跟踪。在6组不同条件下红外视频序列上的实验结果表明,和经典目标跟踪算法相比,本文方法在复杂背景下的平均跟踪成功率提升15%左右,能够有效应对相似虚假目标、遮挡、背景辐射强度变化和探测器晃动等不良因素的影响,适用于复杂背景条件下的红外目标跟踪。
In order to realize robust tracking of infrared target in complicated background with lots of disturbed factors, this paper proposes an infrared target tracking method based on multi-feature correlation filter. Considering the visual attention mechanism and motion mechanism, the spatial feature and motion feature are extracted firstly. Then the multi-feature weighted function is generated by fusing the above two features and the improved convolution feature. Secondly,on the basis of traditional correlation filter, the tracking framework vie weighted correlation filter is presented by introducing multi-feature weighted function which could represent the importances of different candidate regions. Finally, the confidence map which indicates the best target location is computed. The experiments under 6 sequences with different conditions demonstrate that the overage increase of success rate of the proposed method has increased by about 15% compared with other traditional methods, and the proposed method is applicable to infrared target tracking under different conditions efficiently, such as similar alias target, occlusion, thermal radi- ance variation of background and detector motion.
出处
《光电子．激光》
EI
CAS
CSCD
北大核心
2015年第8期1602-1610,共9页
Journal of Optoelectronics·Laser
基金
国家自然科学基金(61102170)资助项目
关键词
红外目标跟踪
多特征融合
相关滤波
权值函数
infrared target tracking
multiple features fusion
correlation filter
weighted function
