基于色度饱和度-角度梯度直方图特征的尺度自适应核相关滤波跟踪

发布时间：2018-08-16 13:20

【摘要】：针对核相关跟踪算法(KCF)对特征敏感及无法跟踪尺度的问题,本文从特征提取和尺度自适应两个方面对核相关滤波跟踪算法进行了研究。提出了一种基于色度饱和度-角度梯度直方图特征的自适应核相关跟踪算法来改善KCF算法的跟踪性能。首先,研究了HSI颜色空间的特点,基于颜色和梯度是互补的图像特征,提出了一种融合了梯度和颜色的HHS-OG特征来有效提高原始KCF算法对目标和背景的判别力。其次,针对KCF无法处理目标尺度变化的问题,在跟踪的检测阶段采用一组固定的尺度因子进行图像块采样,根据得到的滤波响应图估计目标的最优位置和尺度。将所提算法在大量视频序列上进行了跟踪实验,结果显示其平均跟踪速度为37.5frame/s,跟踪精度和成功率分别提升了5.4%和10.1%。实验表明HHS-OG特征具有良好的目标-背景判别能力,能够实现鲁棒跟踪,而尺度自适应策略能较大程度地提高跟踪精度。
[Abstract]:Aiming at the problem that kernel correlation tracking algorithm (KCF) is sensitive to features and unable to track the scale, this paper studies the kernel correlation filter tracking algorithm from two aspects: feature extraction and scale adaptation. An adaptive kernel correlation tracking algorithm based on color saturation-angle gradient histogram feature is proposed to improve the tracking performance of KCF algorithm. Firstly, the characteristics of HSI color space are studied. Based on the fact that color and gradient are complementary image features, a new HHS-OG feature which combines gradient and color is proposed to effectively improve the discriminant power of the original KCF algorithm to the target and background. Secondly, aiming at the problem that KCF can not deal with the change of target scale, a set of fixed scale factors are used for image block sampling in the detection stage of tracking, and the optimal position and scale of the target are estimated according to the obtained filter response graph. Experiments on a large number of video sequences show that the average tracking speed is 37.5frame-1 / s, and the tracking accuracy and success rate are improved by 5.4% and 10.1%, respectively. Experiments show that HHS-OG features have good object-background discriminant ability and can achieve robust tracking, while the scale adaptive strategy can improve the tracking accuracy to a large extent.
【作者单位】：军械工程学院2系;
【基金】：国家自然科学基金资助项目(No.61141009)
【分类号】：TP391.41

【相似文献】