激光与光电子学进展, 2018, 55 (6): 061502, 网络出版: 2018-09-11
目标跟踪中增强梯度阈值的更新方法 下载: 997次
Updating Method of Improved Gradient Threshold in Object Tracking
计算机视觉 目标跟踪 相关滤波器 增强阈值 梯度检测 computer vision object tracking correlation filter enhancement threshold gradient detection
摘要
针对核相关滤波(KCF)算法在每一帧都更新的策略使其不能有效处理目标快速运动及干扰的问题,提出了一种基于增强阈值更新的核相关目标跟踪方法。其在平均峰值相关能量(APCE)的基础上,采用将APCE阈值与APCE梯度阈值相结合的方法来判断跟踪结果的可靠性,以决定模型是否更新。其中将APCE阈值反向加强,APCE梯度阈值正向加强,当APCE和APCE梯度都高于各自阈值时更新,否则停止更新。通过定量及定性实验表明,相对于KCF算法对目标快速运动及干扰等问题的处理,该算法更加有效,提出的以梯度检测跟踪性能及阈值增强的思想对跟踪算法的设计有很好的参考价值。
Abstract
Kernelized correlation filters (KCF) algorithm tracker updating its model parameters at every frame makes it unable to effectively deal with the problems of fast motion and interference of the target in most environments. A nuclear-related object tracking method based on enhanced threshold updating is proposed. Based on the average peak correlation energy (APCE), the APCE threshold and APCE gradient threshold are combined to determine the reliability of the tracking results, which are used to determine whether the model is updated. In this paper, the APCE threshold value is reversely enhanced and the APCE gradient threshold is positively strengthened. When the APCE and APCE gradients are all higher than the respective thresholds, the model is updated, otherwise, the model will stop updating. The quantitative and qualitative experiments show that the algorithm is more effective than the KCF algorithm for the fast motion and interference of the target. The proposed algorithm also provides a good reference value for the design of tracking algorithm based on the idea of gradient detection tracking performance and threshold enhancement.
马晓虹. 目标跟踪中增强梯度阈值的更新方法[J]. 激光与光电子学进展, 2018, 55(6): 061502. Xiaohong Ma. Updating Method of Improved Gradient Threshold in Object Tracking[J]. Laser & Optoelectronics Progress, 2018, 55(6): 061502.