Computer Engineering and Applications ›› 2021, Vol. 57 ›› Issue (2): 198-208.DOI: 10.3778/j.issn.1002-8331.1910-0357

Previous Articles     Next Articles

Improved Target Tracking Algorithm Based on Kernelized Correlation Filter in Complex Scenarios

WANG Bei, CHEN Jinguang, WANG Mingming   

  1. 1.School of Computer Science, Xi’an Polytechnic University, Xi’an 710048, China
    2.Research Institute of Xifang Textile Industrial Innovation in Keqiao, Shaoxing, Zhejiang 312030, China
  • Online:2021-01-15 Published:2021-01-14



  1. 1.西安工程大学 计算机科学学院,西安 710048
    2.柯桥区西纺纺织产业创新研究院,浙江 绍兴 312030


In order to solve the problem of occlusion, scale change and background clutter in visual object tracking, based on the kernelized correlation filter algorithm, the average peak correlation energy occlusion criterion is introduced, an Adaptive fusion Multi-Feature anti-occlusion Kernel Correlation Filter algorithm is proposed (AMFKCF). Firstly, it initializes target features and scale factors. Then, the position and scale filters are trained by extracting and fusing multiple features and scale factors, the center position response of the target is obtained. Finally, according to the occlusion criteria, Kalman filter is introduced to compensate the uncovered and occluded location of the objec center. The AMFKCF and the mainstream tracking algorithm are challenged in CVPR 2013 benchmark. Experimental results show that compared with the mainstream algorithm, the accuracy of the proposed algorithm is improved by 0.115, the success rate is increased by 0.083, the center position error is increased by 14.67 pixels, and the distance accuracy is improved by 9.75 percentage points.

Key words: target tracking, feature fusion, scale adaptation, Kalman filter, Average Peak-to Correlation Energy(APCE), kernelized correlation filter


为解决视觉目标跟踪的遮挡、尺度变化及背景杂波问题,在核相关滤波算法基础上,引入平均峰值相关能量遮挡判据,提出一种自适应融合多特征的抗遮挡核相关滤波算法(AMFKCF)。初始化目标特征及尺度因子,将提取、融合的目标多个特征和尺度因子训练位置和尺度滤波器,得到目标的中心位置响应,根据遮挡判据,引入卡尔曼位置滤波器,对未遮和遮挡的目标中心位置进行补偿。AMFKCF算法与主流算法在CVPR 2013 Benchmark数据集中进行对比,结果表明,AMFKCF算法与主流算法相比精度提高了0.115,成功率提高了0.083,中心位置误差提高了14.67个像素,距离精度提高了9.75个百分点。能够较好地解决遮挡、尺度变化、背景杂波等问题,且兼具核相关滤波算法的优点。

关键词: 目标跟踪, 特征融合, 尺度自适应, 卡尔曼滤波, 平均峰值-相关能量(APCE), 核相关滤波