摘要
针对目标在遮挡、背景杂乱时跟踪定位不准确的问题,提出通道可靠局部秩变换的目标跟踪算法.使用局部秩变换特征结合Lab三通道图像构成16维特征通道,从中选择有效的特征通道进行融合,增强算法对目标特征的表达能力.实验表明,相比于利用可靠性系数选择得到的特征通道,利用经验选择的局部秩变换特征通道在特定场景的跟踪效果更好,对目标的位置估计更加准确.与其他算法进行对比,经验选择方法在特定场景测试上平均速度达到56.7帧/秒,满足实时性要求,在目标测试集上优于对比的两种方法.
To solve the problem of inaccuracy of target location in occlusion and clutter,we propose anobject tracking algorithm based on channel reliable local rank transformation.Local rank transformation features combined with Lab three-channel imagesare used to form16dimensional feature channels,from which effective feature channels are selected for fusion to enhance the expression ability of the algorithm for target features.The experimental results show that,compared with feature channels selected by reliability coefficient,the local rank transformation feature channels selected by experience has better tracking effect in specific scenes and more accurate position estimation of target.Compared with other algorithms,average speed of the empirical selection method is 56.7frames per second in specific scenes tests,which meets real-time requirements,and is better than the two methods in target test sets.
作者
李丽
李均利
田竟民
LI Li;LI Jun-li;TIAN Jing-min(School of Computer Science,Sichuan Normal University,Chengdu 610101,China)
出处
《小型微型计算机系统》
CSCD
北大核心
2022年第5期1081-1087,共7页
Journal of Chinese Computer Systems
基金
国家自然科学基金项目(62002249)资助。
关键词
视觉跟踪
相关滤波
通道可靠性
局部秩变换特征
visual tracking
correlation filter
channel reliability
features of local rank transformation
