摘要
针对传统跟踪方法难以实时准确适应目标尺度变化这一问题,基于核相关滤波跟踪框并采用尺度估计方法,提出一种自适应尺度的目标跟踪算法。对正则化最小二乘分类器进行求解,获得滤波模板,并对候选样本进行检测,估计出目标的位置;利用尺度估计方法,在已确定目标位置处根据前一帧目标的大小对当前帧目标尺度进行检测,由最大的响应值确定当前帧目标的尺度;根据遮挡检测机制,在线更新目标和尺度模型参数。实验结果表明,所提出的算法与其他跟踪算法中的最优者相比,距离精度提高了17.12%,成功率提高了10.77%;在目标发生背景干扰、严重遮挡以及在光照、姿态和尺度变化等复杂场景下,该算法仍具有较好的跟踪效果。
Focusing on the issue that the traditional tracking method is difficult to adapt to the target scale variation in real time accurately, an adaptive scale target tracking algorithm based on kernel correlation filtering tracking framework, which adapts a scale estimation method, is proposed. Firstly, the regularized least squares classifier is used to obtain the filter template, and the position of the target is estimated by detecting the candidate samples. Then, the scale of current frame is determined based on the target size of the previous frame, and the scale samples are obtained by the maximum response value through the scale estimation method. Finally, the target and scale model parameters are updated online according to the occlusion detection mechanism. The experimental results show that the proposed algorithm improves the distance precision by 17.12% and the success rate by 10.77% as compared with the best of the other tracking algorithms. In complex scenes, such as background clutters, severe occlusion, and illumination, posture and scale variation, the proposed algorithm still has a good tracking performance.
作者
高美凤
张晓玄
Gao Meifeng;Zhang Xiaoxuan(Key Laboratory of Advanced Process Control for Light Industry, Ministry of Education Jiangnan University, Wuxi, Jiangsu 214122, China)
出处
《激光与光电子学进展》
CSCD
北大核心
2018年第4期284-290,共7页
Laser & Optoelectronics Progress
基金
国家自然科学基金(61373126)
关键词
机器视觉
目标跟踪
核相关滤波
最小二乘分类器
尺度估计
遮挡检测
machine vision
target tracking
kernel correlation filtering
least squares classifier
scale estimation
occlusion detection
