融合分层卷积特征和尺度自适应核相关滤波器的目标跟踪
发布时间：2018-10-21 11:43
【摘要】：尽管经过多年的研究,尺度变化、形状变化、严重的遮挡、背景干扰、光照变化和相机运动等内外因素引起的目标表观变化,使得目标跟踪仍然是一个极具挑战的问题.为了有效地处理目标表观变化,基于分层卷积特征和尺度自适应核相关滤波器的目标跟踪算法,将目标跟踪分解为目标位置的预测和尺度的估计两个步骤.在目标位置估计方面,区别于传统的基于手工设计特征的目标跟踪算法,我们使用基于分层卷积特征的相关滤波器算法计算出不同卷积层上的跟踪结果置信图,对各个层上得到的结果进行加权求和得到目标置信图,估计出目标的最终位置.在目标的尺度估计方面,为了有效捕捉目标尺度变化,我们首先使用尺度金字塔对下一帧适用的尺度进行预测,同时对目标尺度进行更新.在标准测试集(OTB-50)上的实验结果表明,本文所提出的融合分层卷积特征和尺度自适应的相关滤波器的目标跟踪算法取得较好的精度和鲁棒性.
[Abstract]:After years of research, target tracking is still a challenging problem due to the external and internal factors such as scale change, shape change, severe occlusion, background interference, illumination change and camera motion. In order to deal with the target apparent change effectively, the target tracking algorithm based on hierarchical convolution feature and scale adaptive kernel correlation filter decomposes the target tracking into two steps: prediction of target location and estimation of scale. In the aspect of target location estimation, different from the traditional target tracking algorithm based on manual design features, we use the correlation filter algorithm based on hierarchical convolution feature to calculate the confidence chart of tracking results on different convolution layers. The final position of the target is estimated by weighted summation of the results at each level. In the aspect of target scale estimation, in order to capture the change of target scale effectively, we first use the scale pyramid to predict the scale applicable to the next frame and update the target scale at the same time. The experimental results on the standard test set (OTB-50) show that the proposed target tracking algorithm based on hierarchical convolution feature and scale adaptive correlation filter achieves good accuracy and robustness.
【作者单位】： 华侨大学计算机科学与技术学院;华侨大学计算机视觉与模式识别重点实验室;
【基金】：华侨大学研究生科研创新能力培育计划项目(1511314014)资助 国家自然科学基金面上项目(61572205)资助 福建省自然科学基金项目(2015J01257)资助 华侨大学科技创新能力提升“中青年教师科技创新”计划项目(ZQN-PY210)资助
【分类号】：TN713;TP391.41
，
本文编号：2284991
[Abstract]:After years of research, target tracking is still a challenging problem due to the external and internal factors such as scale change, shape change, severe occlusion, background interference, illumination change and camera motion. In order to deal with the target apparent change effectively, the target tracking algorithm based on hierarchical convolution feature and scale adaptive kernel correlation filter decomposes the target tracking into two steps: prediction of target location and estimation of scale. In the aspect of target location estimation, different from the traditional target tracking algorithm based on manual design features, we use the correlation filter algorithm based on hierarchical convolution feature to calculate the confidence chart of tracking results on different convolution layers. The final position of the target is estimated by weighted summation of the results at each level. In the aspect of target scale estimation, in order to capture the change of target scale effectively, we first use the scale pyramid to predict the scale applicable to the next frame and update the target scale at the same time. The experimental results on the standard test set (OTB-50) show that the proposed target tracking algorithm based on hierarchical convolution feature and scale adaptive correlation filter achieves good accuracy and robustness.
【作者单位】： 华侨大学计算机科学与技术学院;华侨大学计算机视觉与模式识别重点实验室;
【基金】：华侨大学研究生科研创新能力培育计划项目(1511314014)资助 国家自然科学基金面上项目(61572205)资助 福建省自然科学基金项目(2015J01257)资助 华侨大学科技创新能力提升“中青年教师科技创新”计划项目(ZQN-PY210)资助
【分类号】：TN713;TP391.41
，
本文编号：2284991
本文链接：https://www.wllwen.com/kejilunwen/dianzigongchenglunwen/2284991.html
