基于不变矩卡尔曼预测的核相关跟踪算法改进  

Improved Kernel Correlation Tracking Algorithm Based on Invariant Moments and Kalman Prediction

作　　者：洪柱 周论 王科 于航 王振[1] HONG Zhu;ZHOU Lun;WANG Ke;YU Hang;WANG Zhen(School of Electrical and Information Engineering,Wuhan Institute of Technology,Wuhan 430205)

机构地区：[1]武汉工程大学电气信息学院,武汉430205

出　　处：《计算机与数字工程》2024年第7期1944-1949,共6页Computer & Digital Engineering

摘　　要：针对KCF(核相关滤波器)算法在目标发生严重遮挡时易出现目标丢失的问题,提出了一种结合不变距特征及卡尔曼预测的核相关滤波跟踪算法。利用模板图像和当前帧图像的不变矩特征计算相关量,然后根据不同遮挡情况下相关量值的变化情况设定目标遮挡判断机制。当目标未遮挡时KCF算法对目标继续跟踪;当目标被遮挡时利用卡尔曼滤波预测目标位置,目标复现后KCF算法利用预测的位置信息继续跟踪。最后采用OTB-2015数据集测试算法的有效性,实验结果表明,论文算法相对于KCF算法跟踪精度和跟踪成功率分别提高了12.95%和13.28%,有效改善了遮挡情况下的目标跟踪性能。Aiming at the problem that the KCF(Kernel Correlation Filter)algorithm is prone to losing target when the target is severely occluded,a Kernel Correlation Filter tracking algorithm combining the invariant moment features and Kalman prediction is proposed.The similarity metric value is calculated by using the invariant moment features of the template image and the current frame image,and then the target occlusion judgment mechanism is set according to the change of the similarity metric value under different occlusion conditions.When the target is not occluded,the KCF algorithm continues to track the target.When the target is occluded,the Kalman filter is used to predict the target position,and the KCF algorithm uses the predicted position information to continue tracking after the target recurs.Finally,the OTB-2015 dataset is used to test the effectiveness of the algorithm.The experi⁃mental results show that the tracking accuracy and tracking success rate of the proposed algorithm are improved by 12.95%and 13.28%,respectively,compared with the KCF algorithm,which effectively improves the target tracking performance under occlu⁃sion conditions.

关 键 词：目标跟踪 相关滤波 卡尔曼滤波 不变距特征 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]