改进的核相关滤波器目标跟踪算法

针对传统单目标的核相关滤波器(KCF)跟踪算法在目标尺度变化的跟踪中存在的问题,提出了一种基于相关滤波器(CF)和尺度金字塔的多尺度核相关滤波器(SKCF)跟踪算法。首先通过传统KCF跟踪算法中分类器的响应计算当前目标是否受到遮挡,在未受到遮挡的情况下,对当前目标建立尺度金字塔;然后通过相关滤波器求取尺度金字塔的最大响应得到当前目标尺度信息;最后使用新目标图像为训练样本更新目标的外观模型和尺度模型。与核化的结构化输出(Struck)算法、KCF算法、跟踪-学习-检测(TLD)算法和多示例学习(MIL)算法进行对比,实验结果表明,所提出的多尺度核相关滤波器(SKCF)跟踪算法在五种算法中精确度和重合度都取到最高值。所提算法能够广泛应用于目标跟踪领域,对目标进行准确的跟踪。

Improved target tracking algorithm based on kernelized correlation filter

Focusing on the issue that the Kernelized Correlation Filter (KCF) tracking algorithm has poor performance in handling scale-variant target, a multi-scale tracking algorithm called Scale-KCF (SKCF) based on Correlation Filter (CF) and multi-scale image pyramid was proposed. Firstly, the occlusion status of the target was got through the response of the conventional KCF algorithm's classifier. The multi-scale image pyramid was built for the occluded target. Secondly, the scale information of the target was obtained by calculating the correlation filter's maximum response on the multi-scale image pyramid. Finally, the appearance model and the scale model of the target were updated with the fresh target. The experimental results on comparison with some state-of-the-art trackers such as Structured Output tracking with kernel (Struck), KCF, Tracking-Learning-Detection (TLD) and Multiple Instance Learning (MIL) demonstrate that the proposed tracker of SKCF achieves the best accuracy and overlap rate than other algorithms. Meanwhile, the proposed tracker can be widely used in target tracking and achieve high precise target tracking.